KG vs RAG

• First, we solved with Prompt
• Next, we solved with RAG
• Next, we did Summary, RAG
• Next, we moved to Txt2SQL

There is some leftover space for Graph

Build a product and use tech according to needs, No Forefit in the equation :)

Keep Exploring!!!

Learning is Non-Linear

Some days are Sigmoid. Some days are Relu. Novelty is not inventing new stuff but stitching the right techniques in the right proportion.

A few more I relate to my work style :)

• Persist until you figure things out, digging deeper and deeper into unfamiliar systems until it finally clicks, and you understand how they work.
• Have a broad understanding and creative use of Algos, SQL, LLMs, etc. Teach until you get first principles right
• Have the intellectual curiosity to learn about the many new areas you’ll be taken into. Failures today are arsenal for tomorrow.
• Have a willingness to lean into uncertainty. It's okay if a bad day happens.
• Magnitude improvements and novel capabilities. Keep collecting perspectives.
• You will face unique engineering and AI challenges that will make a meaningful impact. To be authentic, you have to have a lot of scars.
• A focus on outcomes, not time-tracking. Yeah, sometimes easy looks complex, complex looks easy.
• The database is always important. Are not afraid of legacy justice systems and would rather make them work than give up.

Keep Going!!!

Txt2SQL

Txt2SQL is easier in straightforward examples, Real database has a ton of complications

Example-

Columns can be generic, Attribute1, Attribute2, We may use Attribute1 for key, Attribute2 for Value. A ton of learning working on it, still trying to get a hold :)

• Hints for clause level
• Hints for join 
• Verbose hints
• Examples in database specific formats
• Categorize on type of errors - Syntax, Missing columns, Missing right joins

Keep Exploring!!!

LLM perspectives

• LLM - Fast, Simple, and Dumb Sometimes
• Prompts - Looks simple but when structured relevant to context can give magical results
• With extended context length, embedding it looks more magical with the abstraction of representations of useful knowledge

LLM generation kids / learning using LLM products will have a different perspective of thinking / before and after ChatGPT :)

Experimentation, Always something to find a solution :)

Some questions and answers take days or weeks, and sometimes the approach moves from LLM to NLP, It's a blend of techniques to make things work.

• How do we optimize RAG with internal documents, Original vs Summary vs Intents, What works best? 
• How do we merge external data? How can we keep versions and relevance?
• More than LLM work, The heavy lifting is for Data preprocessing/cleaning / Embedding on summary 
• When to use LLM vs Multimodals?
• What is the benchmark for our domain and how much do we meet it consistently?
• The transition for LLM, LLM+KG, Creating the data mapping..

A lot of challenges but one at a time, Balancing Consistency, Accuracy, and Latency. If you want to solve real problems you can connect/explore potential learning experimentation opportunities / dedicate some learning hours. Please drop a note to career@proplens.ai

#learnings #NLP #Datascience #RAG #LLMs #perspectives #Datascience 

Keep Exploring!!!

RAG and Prompts - Learning Evolves

In one particular use case, it's a constant process of experimentation and iterations.

• Step #1 - Let me try with prompts - It works but not consistently
• Step #2 - Let me try with a vision prompt - It takes time. 
• Step #3 - Let me merge everything in the database and check. 
• Step #4 - Routing takes 2 seconds, querying takes 2 seconds, and prompting takes 2 seconds. 
• Step #5 - Let's keep accuracy and latency separate, divide everything into separate tracks, and sort out the basics.

Some successes, some lessons, and some learning.

Keep Exploring!!!

Success and Failures in Pushing Ideas in Startups vs Matured Companies

Over the past 4 months, I've been working with really small teams, and the difference in communication dynamics compared to larger teams has been striking.

In my previous roles in product and consulting, I achieved success but with a considerable amount of effort spent convincing and negotiating with numerous people. Here are some of the specific challenges faced when pushing ideas in larger, more mature companies compared to nimble startups:

Increased Lines of Communication: With more team members in mature companies, ensuring everyone is on the same page becomes significantly harder. There's a higher risk of changes in approach, iterations, feedback, and information being lost or mistranslated as it travels through various levels. In contrast, startups often have flatter structures, making communication more direct and less prone to distortion.

Slower Decision-Making Processes: Larger teams often have more layers of approval, which can slow down decision-making. Every stakeholder has their own priorities and concerns, adding to the complexity. Startups, with their smaller teams, can often make decisions more quickly, which allows for faster iterations and innovation.

Greater Need for Consensus: In smaller teams typical of startups, reaching a consensus or getting buy-in for new ideas is often easier. Larger teams in mature companies require more effort to align everyone's visions and goals. This can lead to lengthy discussions and compromises, which may dilute the original idea.

More Stakeholders to Convince: Larger teams come with more stakeholders, each with their own perspectives and interests. This multiplicity can make it challenging to get everyone on board with a new idea. Startups, on the other hand, usually have fewer stakeholders, and the founders or key decision-makers are more accessible, simplifying the process of getting buy-in.

However, the journey you take, whether in a startup or a mature company, will reward you for the risks and decisions you choose to travel with. Each environment has its own set of challenges and rewards, but understanding these dynamics can help in navigating them more effectively.

Keep Going!!!