My Ideal work - Next Plans - Solopreneur

Work as Solopreneur

• Collaborate/partner than build/learn all at once
• Focus on leveraging domain knowledge / ML / AI
• Build strategy  / solution / Translating needs to technical and low-level use cases/solutions
• Leverage existing data / suggest additions / find gaps in data collected/set data clarity
• Advocate tech/business solutions based on all the above aspects
• Teach / Learn / Take care of family/work
• You cannot win every race but you can win only those you take part :)

My Strengths

• Domain - Retail, Supply Chain, Warranty, Offline, Online, Planogram, Refurbishment, Cycle Counting, Warranry, Third Party Integration
• ML - Forecasting, Recommendations
• Vision - Vision for Beauty, Tryon, Smart Retail, AmazonGo-type solutions
• GenAI - Use cases by Text / Vision mapped to domains
• NLP - Teaching gave basics / first principles, NLP, POS Tagging, NER, Transformers, BERT, Spacy, Now OpenAI embeddings
• Databases - Heart and Soul is SQL Server, One default skill is TSQL Developer. Managed BI / SSIS on a need basis
• Professions services/customizations/consulting
• I truly don't believe in I know it all but start from scratch to get to know what's needed :)
• Just keep Learning and Keep Going!!!

Every effort takes time :), I truly cherish all my previous work

• Supply Chain - Xbox Days, Still remember all the tables/flows :)
• Amazon - Features / Ecom / Markets / Ship Fast / Ship More
• Tyco - Retail - Instore - RFID - Replenishment 
• Consulting - Tata Sky, Cavin Care, Govt Depts, Volkswagen, Social Media, Consulting
• Current Role - Applied every other experience to derive value with Data Science

Warranty rewrite took 4 years, Worked on Fixes for two years but it only helped to know drawbacks/business vs tech needs vs migration strategy 

The Retail Architecture rewrite at Tyco was best to Fit how we need to adopt

Patent helped to map new tech vs business process

You can stop earning, but can't stop learning :) :) :)

Take a break focus on a few things, Earn less but Focus more. For broader impact need to have focused plans :)

The key to remain Employable is 

• Learning a lot about things that matter
• Working on problems that interest you
• Make some money with focused work/teaching
• With people you like and respect (this is incidentally how you get co-founders at the same time as the idea)
• Build basic minimal-income scheme, Don't worry continue to evaluate

More Reads - Link

Keep Exploring!!!

Skills = Expertise = Iterations

A research paper takes several months and experiments to evaluate, validate and come out with recommendations

• Ideas may not work the first time and need iterations
• What data says from how much you know the domain determines how well you can interpret data
• The ability to find a quick solution does not mean you understand all aspects of the problem
• I am a firm believer in first principles basics of data - domain - business - technology
• You need to have them in the right proportions to build what the business needs
• Great code does not mean a great solution :)

Keep Exploring!!!

Consulting - People Relationships

Consulting - People Relationships

• Be transparent of skills - Do not cook up what you don't possess
• Document what you know, Be aligned with assumptions
• Frame better and be clear on the formula
• When you do not know something, Provide context with some prework than blanket statements
• Know when to disagree with respect, When to ask for help, and When to provide help
• Apologize if you made a mistake and call out how you avoid it
• When dealing with touch people be slow before you try to defend
• No matter what some people will hate you for no reason, Just don't focus on them

What makes a Team Player

• Not a red flag raiser every time
• Be clear-cut in your deliverables
• Ability to prepare/ask questions upfront
• Proactive to connect dots 
• Think with clarity or take time to connect the dots
• Take vacation but ensure people don't get blocked in your absence
• Focus on few things than faking everything

Knowledge vs Experience

• I always use to say common sense and domain knowledge helps me more than ML out of box techniques
• Leveraging any out of box techniques is good to solve the problem but being able to communicate to business is key
• I would pick a domain + common sense solution before applying any technique which I cannot justify in business-acceptable discussions
• Consulting makes you take multiple roles - product manager - data engineer - tester - program manager - developer - ML engineer endless :) :) 
• Goal-based not task-based, Ability to foresee what others don't see makes an experienced engineer :)

An incomplete list of skills senior engineers need, beyond coding

• How to run a meeting, and no, being the person who talks the most in the meeting is not the same thing as running it
• How to write a design doc, take feedback, and drive it to resolution, in a reasonable period of time
• How to explain a technical concept behind closed doors to a senior person too embarrassed to openly admit that they don’t understand it
• How to lead a project even though you don’t manage any of the people working on the project
• How to give up your baby, that project that you built into something great, so you can do something else
• How to craft a project proposal, socialize it, and get buy-in to execute it
• How to tell someone they’re wrong without making them feel bad
• How to pick your battles

Keep Going!!!

ChatGPT to explain more on company persona - possible use cases

What are machine learning use cases for below company, nature of business explained below. List key ML use cases

<EarnIn lets you access your money as you earn it — not days or weeks later. We’re reimagining the way money moves to make financial momentum accessible to everyone. That starts with payday every day and builds with tools like credit monitoring, automated savings accounts, and low-balance protection. All with no interest, no credit checks, and no mandatory fees. So our customers have as many options as possible to save and spend on their terms.

EarnIn’s pioneering Earned Wage Access app is backed by world-class partners like A16Z, Matrix Partners, and DST. Since our founding in 2013, 3.8 million customers have given it over 380,000 5-star reviews for helping them access over $15 billion in earnings. That means:

- 49% can now cover an unexpected expense of $400 since using EarnIn

- 73% can budget more effectively with EarnIn

- 66% have been able to pay bills on time with EarnIn’s help

We’re proud of that impact. Because no one should feel stuck — held back from their ambitions, their life — simply because they can’t access the money they’ve already earned. That’s why EarnIn is here, to help people move forward with the lives they want today and the goals they have for tomorrow.>

1. Predictive Analysis: Machine Learning algorithms can be used to analyze spending patterns of users to predict future financial needs. This can help to provide guidance on spending habits and saving opportunities.
2. Fraud Detection: Machine Learning can help EarnIn detect unusual patterns of behavior that suggest fraudulent activities, such as account takeovers, fake account creation, or identity theft, protecting the company and its users.
3. Risk Assessment: Machine Learning can identify high-risk customers based on spending patterns, saving trends, and overall financial behavior. This could help the company in making better decisions and managing risks.
4. Customer Segmentation: Through Machine Learning algorithms, EarnIn can group customers with similar financial behavior and design customized solutions for each segment, therefore enhancing customer satisfaction and retaining more customers.
5. Credit Monitoring: Machine Learning algorithms could aid in monitoring users' credit history and providing suggestions to improve their credit score.
6. Personalized Marketing: ML can analyze users' behaviors, preferences, and needs to deliver tailored marketing messages, offers, and recommendations. 
7. Chatbot and Customer Service: A machine learning model can be developed to provide customer service in form of a chatbot. It helps answer frequently asked questions and guide users in utilizing the app effectively. 
8. Customer Churn Prediction: Machine Learning can analyze customer behavior and predict which customers are likely to leave, allowing EarnIn to take proactive steps to improve customer retention.
9. Loan Qualification Prediction: Machine Learning could be used to predict which users would qualify for a loan based on their financial habits, helping users who may not have access to traditional credit.
10. Optimizing loan amounts: Using ML, EarnIn can analyze a user's income, spending habits, and financial obligations to determine the optimal loan amount that they are likely to repay on time without causing financial strain. 
11. Enhancing User Experience: By analyzing user behavior and interactions with the app, machine learning can help EarnIn to continually enhance and personalize the user experience.

What are machine learning use cases for below company, nature of business explained below. List key ML use cases using semi-supervised segmentation and reinforcement learning. 

• Semi-Supervised Anomaly Detection: In credit monitoring, Machine Learning techniques, specifically semi-supervised learning, can be used to detect anomalous transactions that may indicate potential fraudulent activities or risky financial behavior.
• Low-balance Protection: Reinforcement Learning can be used to optimize the timing and amount of alerts sent to users about their low balance, based on their past responses and behaviors. 
• Real-time Credit Score Prediction: Machine learning algorithms, specifically semi-supervised learning, can be used to predict changes in credit scores based on a user’s financial behavior in real-time. This can help users understand how their financial behavior impacts their creditworthiness

What are machine learning use cases for below company, nature of business explained below. List key ML use cases using semi-supervised segmentation and reinforcement learning and datasets

Give an example of  semi-supervised segmentation in NLP

In NLP context, for example, it can be used in text categorization tasks. Suppose a large dataset of customer reviews has to be categorized into topics. We might not have all the manual labels needed to train a fully supervised machine learning model.

We could begin by manually labeling a small subset of the reviews with categories such as "Customer Service", "Product Quality", "Price", etc. This labeled subset can then be used to train a classifier in a supervised manner. 

Then, we can use the classifiers to predict labels for a large set of unlabeled data. These new labeled data can be verified or corrected by humans if necessary, then added to the original training data, creating a larger labeled dataset for the classifiers to learn from. This iterative process continues until the classifier's performance plateaus or meets our desired level--a form of active learning which is a semi-supervised learning approach. This is an example of semi-supervised segmentation in NLP.

Give an example of  reinforcement learning in fintech sector

Reinforcement learning in the fintech sector can be seen in the area of algorithmic trading. In this scenario, a trading algorithm operates in a dynamic environment (the stock market) and makes a series of actions (buy, sell, hold) based on the state of the environment. 

The model gets "rewarded" or "penalized" based on the returns from its actions and over time, through trial and error, it learns the optimal policy - that is, the best actions to take in different market conditions to maximize profit or minimize loss. Using this method, the trading algorithm can continually adapt and improve its strategy as it gains more experience and as the market evolves.

Give an example of  reinforcement learning in fintech sector with simple python code and example

Keep Prompting!!!

ChatGPT

• Decoder-only model.
• Decoder-only architecture does not have an explicit encoder to summarize / context vector the input information
• In Decode only, input sequence is directly fed into the decoder, which generates the output sequence by attending to the input sequence through self-attention mechanisms.

Ref - Link1, Link2

Transformer Key blocks

• In the attention step, words “look around” for other words that have relevant context and share information with one another.
• In the feed-forward step, each word “thinks about” information gathered in previous attention steps and tries to predict the next word.

Ref - Link

What we know about transformers

"What differentiates the Transformer from its predecessors is it’s ability to learn the contextual relationship of values within a sequence through a mechanism called self-attention.

Transformers can be generally categorized into one of three categories:

- encoder onlya la BERT,

- decoder only a la GPT and

- having an encoder-decoder architecture a la T5

Ref - Link

Keep Exploring!!!

Cashflow forecasting

The amount payable depends on 

• Open invoices
• invoices paid
• amount receivables
• number of pending payments

Suggest some more features for cashflow payables forecasting

• Payment due dates
• Early payment discounts
• Seasonal business fluctuations
• Expected returns or refunds
• Scheduled debt repayments
• Forecasted operational expenses
• Unexpected expenditure 
• Tax liabilities
• Upcoming equipment purchase or upgrade costs
• Pending legal disputes or potential fines
• Currency exchange rates (for international payables)
• Credit terms with suppliers
• Inventory purchases
• Projected sales volume
• Market trends and economic conditions
• Inflation rates
• Interest rates on loans
• Business expansion plans
• Capital expenditure for projects.

Suggest data insights to check for Account payables in cashflow forecasting

• Track Trends: Analyze historical data to identify trends in payment cycles, the average amount of payables, the period of payments etc. This can help in forecasting future outflow accurately. 
• Supplier Analysis: Identify the major suppliers and the percentage of total payables they hold. This can help in better supplier management and negotiation for payment terms. 
• Delayed Payments: Analyze the instances of delayed payments, their reasons, and their impact on the company's relationship with suppliers and its credit score. This can help in minimizing such instances in future.
• Cash Conversion Cycle: Check how long it takes for the company's payables to get converted into cash (also known as the payable turnover ratio). If this cycle is high, it may indicate inefficiencies in the company's accounts payable process.
• Impact of Seasonality: Determine if there is any seasonality effect on the account payables of the company. For instance, certain companies might have higher payables during festive season due to increased inventory purchase.
• Default Risk Analysis: Evaluate the probability of the company defaulting on its payables due to insufficient cash inflow. 
• Comparisons: Compare the company's accounts payable turnover ratio with competitors’ or industry averages to assess how

Keep Exploring!!!!

Practical AI 012a: Training an Image Classification Model with Vertex AI

Practical AI 012a: Training an Image Classification Model with Vertex AI

Step 1 - Vertex AI - Data - Datasets - Image Classification

Step 2 - Data Upload

Step 3 - Model Train

Keep Exploring!!!

Quora LLM on the way

 

Quora has many answers, the StackOverflow equivalent for many lifestyles, relationships, travel, and topics across subjects. Today they updated their privacy. Looks like Quora will start using the answers to build an LLM :)

Keep Exploring!!!

coreferences with GPT4

#Coreference resolution (CR) is a task in natural language processing (NLP) that identifies when two or more expressions in a text refer to the same entity. This can include pronouns like "he" or "she," definite nouns like "the car," and other expressions. 

#Coreference resolution helps in other NLP tasks such as machine translation, text summarization, and information extraction. It is required for understanding contextual information and improving comprehension of the text for machines.

Prompt Example

Extract coreferences from two sentences, and prepare the second sentence with coreference substituted. Here are two sentences. Return only second sentence with coreference

1. <First sentence - Steve jobs is great innovator. He founded apple>

2. <Second sentence - What are his key achievements>

Result

<Second sentence - What are Steve Jobs' key achievements>

Keep Exploring!!!

Work = Creativity = Passion

There is no schedule to be happy sometime and sad sometimes. Until we get ideas we keep searching around us. IT / Data science / GenAI all work needs some perspectives / some solutions / some ideas. 

I usually work on two to three tasks at a time and keep a list of things to do.. 

• Task1 - Progress - Pause on Emptiness
• Pick Task 2 - Progress - Pause when not complete
• Read / Get Some ideas 
• Pick Task 1 / 2 based on deadlines 
• Read / Try / Search for ideas / Again Experiment / Endless Cycles :)

Every day you got to observe so much of new things. It is impossible to continue to work in the industry if you do not pick and plan your pace of work

Be your own company, Be your our master, Be your own mentor

There can be 100 painters but only one Leonardo Davinci.It takes time, Follow your pace and consistency.

Keep Going!!!

Data Insights Patterns

1. Descriptive Analysis Pattern: This pattern highlights what has happened and helps in understanding the historical data. This includes defining, classifying and illustrating data trends, summaries, and comparisons. Tools such as dashboards, visual analytics, and reporting can be implemented for better understanding and decision making.

2. Diagnostic Analysis Pattern: This pattern deals with finding out why something happened. It involves more in-depth data exploration, data discovery, correlations, and cause-effect analysis. Advanced tools like data mining, correlations, data discovery tools, etc. can be used in this pattern.

3. Predictive Analysis Pattern: With this pattern, future outcomes are forecasted based on historical data patterns and trends. Techniques such as machine learning, regression models, and forecasting models are commonly used in this pattern. The predictive analysis helps in making proactive decisions and risk mitigation.

4. Prescriptive Analysis Pattern: This pattern provides insights on what action to take to eliminate a future problem or take full advantage of a promising trend. Optimization, simulation, decision tree, and rules-based system can be used for this type of analysis. It advises on possible outcomes and businesses can prepare in advance, using this type of analysis.

5. Data Discovery Pattern: This pattern helps in finding hidden patterns and unidentified trends in the data. This pattern may employ data mining, data visualization, and big data analytics tools to enable analysts to find useful insights.

6. Anomaly Detection Pattern: The process of identifying unusual patterns that do not conform to expected behavior, called anomalies. Machine learning algorithms and statistical techniques are used in this pattern to detect outliers or anomalies in the data.

7. Text Mining Pattern: This pattern is focused on extracting valuable information from textual data. It may involve techniques such as natural language processing (NLP), sentiment analysis and text analytics to derive insights from unstructured text data.

8. Social Media Analysis Pattern: This pattern involves analyzing data from social media platforms to gain insights about user behavior, trends, and sentiments. Social network analysis, sentiment analysis, and trend analysis are common techniques used in this pattern.

9. Visual Analytics Pattern: This pattern uses visual representation techniques to present data and insights and make them easily comprehensible. Data visualization tools such as charts, graphs, infographics, and dashboards are used.

10. Geospatial Analytics Pattern: This pattern involves analyzing data based on geographical or spatial factors. Geospatial analysis may include using Geographic Information Systems (GIS) and location intelligence

Keep Exploring!!!

Negotiation for Life :)

Text extracted with anthiago

Talk #1

Margaret Neale: Negotiation: Getting What You Want

Extract key summary as oneliners. Each line with a max length upto 15 words. Summary in below format

1.

2.

3.

• Negotiation can improve one's career and life significantly.
• Negotiation should be viewed as a problem-solving approach.
• The goal of a negotiation should be a good deal.
• It's crucial to know alternatives, reservation price, and aspiration.
• Four steps of negotiation include assess, prepare, ask, package.
• Understanding the counterpart's interests is crucial during a negotiation.
• Bringing unique information allows you to negotiate effectively.
• Proper planning brings successful negotiation outcomes.
• Practicing negotiation in lower-risk situations builds confidence.
• Learning from both successful and failed negotiations is important.
• Negotiation is seen as a tool to create value for both parties.
• Continuous practice leads to expertise in negotiation.

Talk #2

"Negotiating with Emotion" with HBS Online Professor Mike Wheeler

• Observing team dynamics can give insight in negotiations.
• Non-verbal signals can reveal truth in negotiations.
• Preparation also involves being in the right emotional state for negotiation.
• Successful negotiators are both calm and alert.
• Preemptively identifying personal triggers can aid in negotiation.
• Emotional preparation in negotiation should not be skipped.
• Skill of being assertive and amicable simultaneously exists.
• Confidence at the negotiation table can come from personal abilities.
• Emotional expressiveness can prompt reciprocity from the other party.
• In remote negotiations, attempt small talk for better outcomes.
• Naming a problem can rectify it when negotiating under time pressures.
• Larger representation doesn't always equate to larger power.
• Paying attention to process can speed up negotiation.
• Two different negotiation styles can find middle ground by discussing process.

Many things in life need a lot of perseverance, grit, passion, empathy :)

Keep Exploring!!!

2050 Future

 

Keep Imagining!!!

Healthcare and AI

Text extracted with anthiago

Talk #1

What's Next For AI In Healthcare In 2023? - The Medical Futurist

Extract key summary as oneliners. Each line with a max length upto 15 words. Summary in below format

1.

2.

3.

• The FDA has approved many AI healthcare applications.
• The FDA is developing a database for AI-based healthcare tech.
• AI poses a regulatory challenge as it learns and evolves.
• Radiologists now have guidelines to utilize AI Technologies.
• Regulatory bodies are working on governing AI's large language models.
• Misuse of AI could spread misinformation or breach patient privacy.
• Despite challenges, AI's benefits to healthcare far outweigh the difficulties.
• Companies are integrating AI services like GPT into their business.
• AI is predicted to breakthrough in drug discovery this year.
• AI Solutions usage will increase in established areas like oncology.
• AI will start emerging in new fields like mental health.
• AI will lead to increased efficiency and adoption in healthcare.
• AI will help in revealing solutions that were overlooked before. 

Talk #2

Medical Uses of ChatGPT - The Medical Futurist

• GPT is used widely for academic and coding purposes.
• GPT, the latter version, is being studied for medical usage.
• GPT can help in generating and analyzing clinical notes. 
• It can also answer general health-related queries.
• GPT can pass the U.S. medical exam without specialized training.
• It can help doctors and nurses create accurate clinical notes.
• Medical chatbot, a potential use case, can address a doctor shortage.
• GPT's current limitation includes not citing references and sources.
• GPT's utilization in healthcare implies users being fact-checkers.
• Google's DeepMind has released Medpalm to answer healthcare queries.
• As AI improves, risk of lack of healthcare personnel threatens.
• GPT and AI tech will profoundly change the future of healthcare.

Keep Exploring!!!

AI and Product Management

Webinar: How to Be an AI Product Manager by Facebook AI Product Leader, Natalia Burina

Extract key summary of below discussion in one liners as

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

Key Summary

• AI Product Managers need to identify business problems that AI can solve, envision strategic solutions and bring them to life.
• There are different types of AI PMs, including ones focusing on products, platforms, AI research, and building AI responsibly.
• AI shifts software development from a deterministic process to a probabilistic one, making it essential for PMs to understand and manage the trade-offs.
• Some practical skills for AI PMs include: understanding how to use rapid innovation tools, knowing the various categories of problems that machine learning can solve, and comprehending the AI product pipeline.
• AI should be developed responsibly, ensuring it is fair, private, robust, explainable, accountable and reliable.
• AI PMs need to identify the right problems AI can solve, understand the technology's potential, align and track the right business metrics, and understand the potential harms of the technology.
• AI should support business metrics and AI metrics should relate to greater business goals, requiring engagement with all stakeholders to define suitable metrics.
• An AI PM should foster an experimental culture, taking calculated risks and being willing to learn from failures, as AI rewards those willing to do so.
• Barina's tips for success include telling a compelling story, preparing a six-month plan to stay focused, and using Andrew Bosworth's cold start algorithm when starting a new job.

Webinar: AI/ML Product Management by Uber Sr PM, Kai Wang

• Explains how machine learning customizes the Uber experience, such as determining the best driver, pickup location, delivery time, and ensuring transaction safety.
• Talks about AI and machine learning product management, including the definition, types, and the skills required for AI product managers.
• Discusses the differences between AI products and traditional software products in terms of defining success, project and risk management, as well as the needed technical understanding of AI and machine learning.
• Different types of AI products include platforms/frameworks, AI applications addressing specific use cases, and applied machine learning products utilized in daily life such as Google Search, self-driving cars, and digital assistants like Siri.
• According to Kai, 10% of AI product managers work on machine tooling, 20% work on AI services, and the majority focus on applied machine learning.
• Importance of having a fallback plan for when AI models fail was stressed, reminding AI product managers to prepare for wrong predictions.
• Emphasizes remaining user-centric while being technically proficient and understanding the needs and behaviours of users

Panel Discussion on The Future of AI in Product Management

• The panelists highlighted the use of AI in various sectors, including the restaurant industry and medical field.
• They identified the need to correct existing biases in AI data sets to prevent further embedding of such biases.
• They believe AI technology should be accessible and usable across different departments in an organization.
• The panelists suggested product teams should explore AI and understand its potential for solving customer problems and enhancing their work.
• They foresee AI as a service and believe AI integrations will become a substantial part of the tech industry in the next five years.
• The panelists spoke about the need for AI to complement human skills, rather than replace them.
• They cautioned against over-promising on AI capabilities, emphasizing it should be seen as a tool for efficiency and problem solving, not a replacement for human roles.
• The panelists called for a realistic approach to AI adoption, leveraging human strengths alongside AI capabilities.

Keep Exploring!!!

MLOps Notes

Why MLOps

Quicker experimentation cycle -> More models

Quicker productization cycle -> More models in production

Full traceability for all models -> More models in production safely and scalability"

Tools for MLOps

• Data Analysis - Python, Pandas
• Source Control  - Git
• Test & Build Services - PyTest & Make
• Deployment Services - Git, DVC
• Model & Dataset Registry - DVC[aws s3]
• Feature Store - Project code library
• ML Metadata Store - DVC
• ML Pipeline Orchestrator - DVC & Make
• Experimentation Tracking - MLFlow

Mops Stack

GCP MLOPs

• MLOps
• Featurestore
• Model Monitoring

AWS MLOps

• Feature store
• Model Monitoring

Azure MLOps

• Feature store
• Model Monitoring
• Set up MLOps

MLFlow

MLflow - Tracking experiments, Packaging ML code, Managing and deploying models, central model store  

Made with ML

• Made With ML
• CI/CD for Machine Learning
• Model Serving
• Distributed training
• Monitoring Machine Learning Systems

MLOps Courses

• MLOps Zoomcamp
• MLOps-Basics
• MLOps Guide
• awesome-mlops
• dtu_mlops
• MLOps Course
• MLOps Tutorial #1: Intro to Continuous Integration for ML

Kubernetes

• Intro to Kubernetes Workshop
• Kubernetes slides
• Kubernetes The Hard Way
• From Notebook to Pipeline in No Time with LineaPy and DVC - Thomas Fraunholz at PyData 

Keep Exploring!!!

AI and product management

Key talk and Summarized with AI

AI and product management | Marily Nika (Meta, Google)

Extract top key notes as one liners in below format

1.

2. 

3.

Summarized with Chat GPT-4

1. Avoid using AI for the sake of using AI; ensure there's a real problem or pain point that needs to be solved in a smart way with AI.
2. PMs should get comfortable working with research scientists who can develop AI and machine learning models to enhance product features.
3. Don't implement AI into minimum viable products (MVPs); initial product propositions should be validated with users through prototypes before investing in AI.
4. Understand that the quality of your AI product will depend on the diversity and volume of data you have; using the same data set as everyone else will yield similar results.
5. Training a model involves providing the model (or "AI brain") with repeated examples (like images or text) until it learns to recognize and categorize inputs with a reasonable level of accuracy.
6. AI will not replace product managers but rather enhance their capabilities by handling repetitive tasks and providing intelligent insights, allowing them to focus on strategic aspects of product management.
7. There is an increasing need for product managers to become AI product managers who can understand, implement, and effectively leverage AI technologies in creating and managing products.
8. AI product managers need to be prepared for uncertainties, complex leadership scenarios, data sourcing challenges, and different career progression paths.
9. To get buy-in for AI projects, provide evidence of successful AI implementations, propose fallback plans, and show potential to monetize AI capabilities.
10. Continuous learning, keeping updated with advances in AI technology, and educating yourself through resources like online courses, AI platforms like AutoML and OpenAI, and academic research are recommended approaches to becoming a strong AI product manager.
11. Creating and offering courses on AI and other subjects can be rewarding and beneficial, both for sharing knowledge and for personal growth and learning.

Keep Exploring!!!

Keep Building!!!

Career - Next options

• Advisory Role
• Just do nothing - Observe the industry / Experiment ideas
• Build MVP / High-value solutions
• Continue on Teaching
• Write a book recounting highs and lows in Life
• Work for Non-Profit Social cause

Keep Thinking!!!

Senior Roles = Product + Data + ML - All Hats :)

An ideal senior role in my view,

Blend your ideas in DATA + DOMAIN + AI + Business Value to work on use cases. Tech expertise is important, but it is equally important to solve the right problems in the right way. Building a world-class solution requires goal-oriented thinking, not task-based thinking. Chose careers that build your passion, Find something where you can contribute significantly and also get yourself involved in bigger goals.

Keep Exploring!!!

Your weakness is your strength

Talent is not 10 people doing the same 10 things
Talent is 10 people doing the same things in their own way
Experience is not just building things
Experience is knowing how to build a compelling and balancing tech and customer lens
Years of Experience do not mean Experience in multiple problems
10 years in the same projects vs 10 years in 10 projects both different
Titles do not mean expertise
Be better than yourself
Focus on First principles
Learn at your own pace

Keep Going!!!

NLP use cases before and after ChatGPT

 

Many use cases will converge
Many startup ideas will converge
Who has the data is the King!!!

Keep Exploring!!!

Ray Dalio - Perspectives - Summarized with ChatGPT

Ray Dalio - Perspectives - Summarized with ChatGPT

Principles for Success from Ray Dalio: Founder of the World’s Largest Hedge Fund

Generate transcript with anthiago

This model's maximum context length is 8192 tokens. However, your messages resulted in 13145 tokens. Please reduce the length of the messages.

Divide into Several blocks and Summarize

Extract key summary in few lines point by point

1.

2.

3.

Key Lessons

• Ray Dalio proposed life exists in three phases: being dependent on others and learning, working with others depending on you, and finally being free of obligations
• Dalio revealed he significantly benefitted from his failures and realizing that pain plus reflection equals progress.
• Failure helped Dalio reshape his perspective, approaching decision-making with humility and audacity, stress-testing his decisions and seeking people who disagreed with him to stimulate learning and enhance decision making.
• Importance of considering different views within the team and exploring the reasoning behind each one to get a broader understanding.
• Importance of algorithmic decision-making, arguing that it provides the knowledge necessary for efficient decisions
• Attention to the inequality in the education system, recommending a bottom line and private-public partnerships as ways to improve it.
• Providing opportunities and helping people to be more independent and useful
• Creating opportunities for people to be productive and make meaningful contributions to society.

Principles For Success by Ray Dalio (In 30 Minutes)

• Independent thinking and self-awareness stressed.
• Learning from mistakes and evolving
• Success lies in continuous personal evolution, not just in achieving goals.

What are key  five-step decision-making process. Explain in 20 words

1. Identify Goals: Understand your own needs, values and the outcomes you want to achieve.
2. Pinpoint Problems: Determine issues that might prevent you from achieving the goals.
3. Diagnose Problems: Delve into the root causes of these problems by analyzing them without jumping to conclusions.
4. Design Plan: Develop a strategic plan to address and eliminate the root causes of the problems.
5. Execute Plan: Implement the design, continuously learning, adapting and pushing towards the goal

Keep Learning!!!

Reduce your time to Read / Listen / Summarize / Save time

Reduce your time to Read / Listen / Summarize / Save time

1. Pick the video to summarize - Link

2. Get the transcript from - anthiago

3. Use GPT4 to get key summary

Extract top 10 key predictions from below text. Summarize each line in 10 words only

1.

2.

3.

The Results are

• By end of the next year, chat ChatGPT will be available on mobile phones without internet.
• The figure out mechanism is about making humans scale, akin to an 'analyst with a bad memory'.
• The future of AI is pushing internet intelligence to the edge for every person, country, culture, and company.
• Most popular language models in the world, gptj and neon x, are paving the way for deploying AI systems on-premises.
• Stable diffusion technology can reduce 100,000 gigabytes of images to a 2 gigabyte file that can generate anything.
• Transforming language into images and visuals; four of the top 10 App Store apps based on this premise.
• The second-generation model of this technology will produce photorealistic images and video.
• The trajectory indicates that in five years there will be no need for programmers.
• Open-source models will overtake proprietary models as they offer ownership and interpretation advantages.
• By next year, the capacity of chat GPT will increase, enabling the passage of a level 3 programmer exam

Points I tend to agree with, highlighted in a different color

Earlier I used to spend more time reading complete tech talks, and webinars. Now scanning a few mins and the view count convey intent to read. Passed the video link to generate a transcript. Leverage chatGPT to summarize key points. These five key points and my comments added

• Very soon - Edge-optimized offline LLM models ChatGPT will be available on mobile phones without the Internet. - Edge optimized / Like keyboard word suggestion sentence/reply recommendation possible
• Open-source models will overtake proprietary models as they offer ownership and interpretation advantages. - Now we see Google hosting an unlearning competition for models, More models more explainability/reasoning. The VGG moment is crossed. Resnet / RNN / Another transformer moment coming
• The trajectory indicates that in five years there will be no need for programmers. - This is scary. The need to memorize syntax is no longer needed. How we connect the dots, and consider aspects of the solution becomes key
• Stable diffusion technology can reduce 100,000 gigabytes of images to a 2 gigabyte file that can generate anything. - Right now vision / 3D is very early stages. Next / Coming years 1~2 will get realistic / production version quality with more data trained and larger models tuned to prompts

Keep Exploring!!!

Decision tree - Summary

A decision tree is a type of supervised learning algorithm that is mostly used in classification problems but can also be used for regression

Entropy(p) = - p*log2(p) - (1-p)*log2(1-p)

A high entropy value indicates a high degree of disorder or impurity in the data

Gini(p) = 1 - (p^2 + (1-p)^2)

The lower the value of Gini impurity, the better it is.

We look at Entropy/Gini, Basically depending or target class which features help us with good distribution. The goal is to build a tree with the right choices of root, next level upto leaf nodes.

Information Gain is a statistical property that measures how well a given feature separates the training examples according to their target classification. It is calculated by comparing the entropy of the dataset before and after a transformation. The feature with the highest information gain after the split will be chosen as the node. The process is then repeated for further branch nodes.

• Information gain is the decrease in entropy after a dataset is split on an attribute.
• Information Gain measures the reduction in entropy (or impurity) achieved because of the split. 
• In the decision tree algorithm, at each node, the feature that provides the highest Information Gain is chosen for the split.

Keep Exploring!!!

ChatGPT + DALLe Strory writing

Prompt Input

write a simple story of a boy wandering in the woods and enjoying the weather, call out his peace and happiness in a short 100 words story

Story

Once upon a time, a delight-filled boy named Sam found himself wandering through the enchanting woods. Each step he took was a dance with nature, his laughter harmonizing with whispers of the breeze. Lush green leaves rustled softly, waving hello, while the sun winked through gaps, spattering gold onto the forest floor. His heart swelled with peace as he relished the curious symphony of birds, the orchestra of woods. The crisp air filled his lungs, painting a smile only pure happiness could muster. The weather was his best companion, warm yet invigorating. Sam realized, within these woods, he found an unbounded joy.

DALLE Prompt

Once upon a time, a delight-filled boy named Sam found himself wandering through the enchanting woods. 

Delight-filled boy named Sam took was a dance with nature, his laughter harmonizing with whispers of the breeze. Lush green leaves rustled softly, waving hello, while the sun winked through gaps, spattering gold onto the forest floor. 

Delight-filled boy named Sam heart swelled with peace as he relished the curious symphony of birds, the orchestra of woods. The crisp air filled his lungs, painting a smile only pure happiness could muster. The weather was his best companion, warm yet invigorating. Sam realized, within these woods, he found an unbounded joy and happy faces.

Keep Exploring!!!

Vision Interesting Startups

• Virtual Try on Beauty- Perfect365, Orbo
• Virtual Try on Jewelry - Tangiblee Virtual Try-On
• Image Catalog Management - Vue.ai, Spyne.ai, Stylitics, wide eyes
• In store analytics planning solution- Increff, JDA
• Face Beautifier - Banuba

With Generative AI, Vector databases many of this will face tough competition to sustain

Keep Exploring!!!

3.5 years of AI / ML work and some key projects

Some key moments of the past 3.5 years of work 

• Setting up / Driving a Virtual NRF - Driving everything from India
• Strategy roadmaps based on the current state of multiple customers in  Retail, Logistics, Beauty, etc..
• Vision-based products / Roadmaps / Production Architecture - Virtual Tryon
• DAM / Stylitics / Try on Solutions 
• Vision Products / Projects - Skin Care / Leaf-based plant classification
• Forecasting projects - 300K products for Beauty product retailer
• Bundle recommendations - For a clothing retailer, Figuring out what sells well
• 2 months AI + Analysis + Troubleshooting a slow-performing Trading App (Air Crash Investigation type work :))
• Image Search Engine, Image Catalog creation
• A ton of tech reviews, architecture discussions
• Code up as and when needed, Code / Learn / Handle both tech + business audience
• Three batches of training AI / ML for 150+ product managers, Publishing POV
• Reporting feedback/improvements for AzureOpenAI, Google partners, etc.
• Currently in LLM, GenAI Mode
• External Talk in One of the Conferences for Virtual Try on
• Almost all good moments, except a few situations where I would be cautious if I spot such symptoms
• Lot of coding/teaching for Upgrad :) to get better at basics :)
  

In the next 5 years how we code, will it be mostly prompt-based or idea based looks like a lot of new approaches in development :)

• Tyco and Microsoft taught me a lot of domain
• Even after a master's in ML, Domain knowledge + Common sense helps me more than ML views
• Still could recollect key tables at least 100 tables in Reverse Logistics work - Product, Warranty, Msops, SST Tracking, Repair, Warranty etc..
• The warranty rewrite work is still memorable and applicable till today
• The 3PL touch point connectivity is still relevant in supply chain visibility
• Sensormatic gave the RFID + EAS + People counting + Instore retail operations
• The heart and soul of instore operations is based on store planning + planogram + store layout + real-time alerts + cycle counting + a lot of real-time opportunities with Vision
• Now if I look back a lot of ML is applicable and I will rework If I have to redo those problems yet again :), Classifying the type of customer issues in XBOX, using NLP to address customer issues with GenAI, Past had a lot of data
• Now all data can generate signals - Video, Audio, Text on top of RDBMS

Learned a few things (New start)

• Optimization opportunities
• Exploring pyomo / other relevant connected opportunities
• Azure data curate features/masking/removing/compliance
• A ton of training / long term - only to learn more 'persistence'
• A ton of mentoring internship projects 
• Reviewing/panelist in several events
• Tyco was a bit relaxed with work, This is a marathon

Wishlist

• Need a break!!!
• Write a book
• Take a break
• Try building some ideas/products
• Business + Tech is always essential to see the big picture / Explore options / Freelance

My perspective

Ref - Link

Why AI / ML ideas Fail in Large Enterprises?

Keep Thinking!!!

Streamlit Deployment Notes examples

•  30 Days Streamlit learning
• Deploy streamlit to Google App Engine in the flexible environment without any docker/container.

runtime: python
runtime_config:
operating_system: “ubuntu18”
runtime_version: “3.8”
entrypoint: streamlit run --server.port=8080 --server.address=0.0.0.0 --server.enableCORS=false --server.enableWebsocketCompression=false --server.enableXsrfProtection=false --server.headless=true app.py
env: flex
network:
session_affinity: true

• ViLT-GPT
• face_detection_app
• NLP Based App with Streamlit
• Deploy Streamlit using Docker
• Deploy Streamlit using Kubernetes
• Ask my PDF
• Langchain - Text Summarization App
• Unlock your creativity with LLMs

Keep Exploring!!!

Switchon - Vision Solutions

Vision for Automated Inspection 

 

Ref - Link

Keep Exploring!!!

Lightweight Deep Learning - Model Tuning - Model Compression

Key Techniques

• Pruning
• Weight Sharing
• Quantization
• Low-rank Approximation
• Sparse Regularization
• Distillation

Pruning Weights

• Motivated by how real brain learns
• Remove weights which 𝑤𝑒𝑖𝑔ℎ𝑡 < 𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑
• Retrain after pruning weights
• Learn effective connections by iterative pruning
• Between L1 and L2, which regularization is better

Criteria for Pruning 

• Minimum weight - Pruning by the magnitude of Kernel weight (L2 norm)
• Smallest activation - Prune Kernels that lead to feature maps with the, smallest activations

Weight Sharing

• Compress the neural network with weight sharing
• Use a low-cost hash function to randomly group connection weights into hash buckets

Quantization

• Binarize all the weights and activations, turning DNN into Binarized Neural Network (BNN), in order to reduce memory consumption and increase power-efficiency.
• The method binarizes both weights and activations, in contrast to BinaryConnect which binarizes only weights

Low-rank Approximation

• SVD

Sparse Regularization

• Zero-out groups of weights using sparsity-inducing penalty

Distillation

• Based on teacher-student model
• Given trained deep neural network (DNN), called ‘teachers’, make a compressed ‘student’ model, with similar accuracy using quantization and distillation
• Teacher = Original Deep Model, Student = Quantized Model

Ref  - Link

Keep Exploring!!!

Your weakness is also your strength

Appraisal for Horse - You run fast but you don't load as much as an elephant, improve your load capacity 
Appraisal for Elephant - You take heavy loads but you are slower than a horse, We need you to be good at speed
In the end, both of them will become weak :)
Remember you are your own competitor

Keep Running!!!

Docket Notes - Image Sizes

Docker: Reduce the size of a container

using https://alpinelinux.org can reduce the size and deployment time of a container by up to 40%

Install minimum packages

Reducing Docker Image size

• Bulkier images take time to get pulled.
• Increased time to spin up a container using it.
• Include a .dockerignore file.
• Reduce unnecessary dependencies.
• multistage build pattern. Made a binary on the build stage and used it to run the container.

.dockerignore

docker-squash

• Squashing helps with organizing images in logical layers. Instead of having an image with multiple (in almost all cases) unnecessary layers - we can control the structure of the image.

Keep Exploring!!!

Why consulting does not impress me?

• Short cycles with limited data exposure limit understanding of solutions. 
• Innovation needs iterations, and scoped work does not allow for a deep dive to solve the problem at scale. 
• Tech expertise is important, but it is equally important to solve the right problems in the right way. 
• The perspective of domain and business problems with tech solutions often lacks, and quicker cycles are good, but meaningful and impactful solutions are what matter, not intermittent or stop-gap solutions. 
• Building a world-class solution requires goal-oriented thinking, not task-based thinking.

Keep exploring!

Experience = Domain + Data + Tech + Persistence

I could deeply relate to this post. I believe that having a deep understanding of the data and domain knowledge is essential for producing the best results. I have experienced this firsthand in my work, such as rewriting warranty engines, designing solutions to meet the needs of the business, coming up with approaches for warranty corrections, and supporting on-the-fly changes for different markets.

The relevance and completeness of the solution is the result of production support, familiarity with the tables and columns, and an awareness of any data issues.

In order to achieve the best results, it is important to have a comprehensive understanding of the data and domain knowledge. This requires consistency and perseverance, and it is only through this that the big picture can be seen.

Having a thorough knowledge of the data and domain can be difficult to acquire, but the rewards are worth it. Not only will it help to produce better results, but it will also give you a better understanding of the project as a whole. It is also beneficial to have a good understanding of the different markets and the changes that may be required for them.

Having a deep understanding of the data and domain knowledge is essential for producing the best results. It requires consistency and perseverance, and it is only through this that the big picture can be seen. It is important to have a comprehensive understanding of the data and the domain knowledge, as this will help to produce better results and give you a better understanding of the project as a whole. Additionally, it is beneficial to have a good understanding of the different markets and the changes that may be required for them. With the right knowledge and understanding, you will be able to produce the best results for any project.

Learning's as part of current roles

• Discuss with both between business and technical audiences.
• “Hands-on” and “agile” to fix where it breaks
• Technically hand-holding data scientists, engineers, and architects depending on solution needs. Drive back to first principles
• Collaborating with colleagues from different cultures and time-zones across the globe

Keep Exploring!!!

Finance - Learning

What are payables in finance ?

Payables in finance refer to the amounts owed by a company to its creditors, suppliers, or other entities as a result of purchasing goods or services on credit. These are short-term financial obligations that need to be settled within a specific period, usually within a year. Payables are recorded as liabilities on a company's balance sheet and are also known as "accounts payable" or "trade payables."

In simple terms, payables represent the money a company owes to others for goods or services it has received but not yet paid for.

Does financial planning happen with cash in hand or planned cash and estimate with it

Financial planning typically involves both cash in hand (current cash available) and planned cash (future cash inflows and outflows). It is a comprehensive process that helps individuals and businesses to manage their financial resources effectively, achieve their financial goals, and ensure long-term financial stability.

In financial planning, you start by assessing your current financial situation, including cash in hand, assets, liabilities, and income sources. Then, you set financial goals and create a plan to achieve those goals by estimating future cash inflows (such as income, investments, and other sources) and outflows (such as expenses, debt payments, and savings).

The plan should be flexible and adaptable to changes in your financial situation, goals, and market conditions. Regular monitoring and adjustments are essential to ensure that the plan remains effective and aligned with your financial objectives.

In summary, financial planning involves working with both cash in hand and planned cash, as well as making estimates and projections to create a comprehensive and adaptable plan for achieving financial goals.

What are key factors that influence payables in finance ?

Several key factors influence payables in finance. These factors can affect the amount, timing, and management of payables for a company. Some of the key factors include:

Credit terms: The credit terms offered by suppliers, such as payment due dates and discounts for early payments, can significantly impact the amount and timing of payables. More favorable credit terms can lead to lower payables, while less favorable terms can result in higher payables.

Volume of purchases: The volume of goods or services purchased on credit directly affects the level of payables. Higher purchase volumes lead to higher payables, while lower volumes result in lower payables.

Payment policies: A company's payment policies, such as the frequency and timing of payments to suppliers, can influence the level of payables. Companies that delay payments or negotiate longer payment terms with suppliers may have higher payables, while those that pay promptly or take advantage of early payment discounts may have lower payables.

Economic conditions: Economic conditions, such as interest rates, inflation, and overall market conditions, can impact payables. For example, higher interest rates may encourage companies to pay off their payables more quickly to avoid additional interest costs, while lower interest rates may lead to companies taking longer to pay their payables.

Supplier relationships: The quality of relationships with suppliers can influence payables. Companies with strong supplier relationships may be able to negotiate better credit terms, resulting in lower payables. Conversely, strained relationships may lead to less favorable terms and higher payables.

Industry practices: Industry practices and norms can also affect payables. Some industries may have standard payment terms or practices that influence the level of payables for companies operating within that industry.

Cash flow management: Effective cash flow management can help companies optimize their payables by ensuring that they have sufficient cash to meet their obligations while minimizing the cost of carrying payables. Companies with strong cash flow management may have lower payables, while those with poor cash flow management may struggle to meet their payment obligations, resulting in higher payables.

Understanding and managing these factors can help companies optimize their payables, improve cash flow, and maintain strong relationships with suppliers.

Keep Exploring!!!