ESG leadership exploring domain‑specific ESG solutions? Let’s talk.

• Common pain points I hear: fragmented plant/supplier data, manual ESG/BRSR/CSRD spreadsheets, and generic tools that don’t fit textiles/fashion/manufacturing realities.
• If you’re a CSO/Head of ESG, CFO/CIO for ESG data & systems, or an ESG/impact investor interested in sector‑specific ESG infra, I’d love to connect, understand your current challenges, and explore a low‑risk pilot or partnership.
• phygitalytics.com is actively working in this space.

Let’s talk.

Vector Databases Reads

 Milvus Notes - Index / consistency / availability options

#1. Index type - usecase

• IVF_FLAT - High-speed query
• IVF_PQ - Very high-speed query
• HNSW - High-speed query

Inverted File (IVF): An IVF index divides the vector space into several clusters and holds an inverted file for each cluster, recording which vectors belong to the cluster.

IVF Flat: This is a combination of IVF and flat index. It uses the IVF index to partition the data into clusters and then uses the flat index (brute-force search) within each cluster.

Hierarchical Navigable Small World (HNSW): HNSW builds a multi-layer navigation graph to represent the vector space.

#2. Consistency levels - Strong, Bounded, Session or Eventually

• Strong - Most strict
• Bounded staleness - allows data inconsistency during a certain period of time.
• Session - Like dirty reads 
• Eventually - weakest level among the four.

#3. HA - In-memory replicas help Milvus recover faster if a query node crashes.

#4. Vector search & Hybrid Search params offset, limit

offset - Number of results to skip in the returned set

limit - Number of the most similar results to return

How indexing and querying works

• Trees – ANNOY -  Annoy (Approximate Nearest Neighbors Oh Yeah)
• Proximity graphs - HNSW Hierarchical Navigable Small World (HNSW) Graphs
• Clustering - FAISS
• Hashing -  LSH - Locality-Sensitive Hashing (LSH)
• Vector compression - PQ or SCANN. - ScaNN (Scalable Nearest Neighbors). Product Quantization (PQ): PQ index compresses vectors into compact codes and is beneficial for large-scale, high-dimensional data.

More Reads

• Utilizing Few-shot and Zero-shot learning with OpenAI embeddings
• Query Comparison
• Accelerating Similarity Search on Really Big Data with Vector Indexing

Keep Exploring!!!

Phygitalytics

 

Welcome to Phygitalytics!!!

Thank you to all past, present, and future customers!

Happy Responsible AI Adoption!!!

🧠 The Silent Killer of AI Adoption: Leadership-Level AI Illiteracy

The bigger problem of AI adoption isn't tooling or data, it's ‘AI illiteracy’ at the leadership level.

The kind of questions we ask reflects how much we know. And too often, the questions reveal a dangerous gap:

1️⃣ Misunderstanding ML like software engineering

“I will give you 10 samples, can you build a model?”
“You have trained a model on this data, why can't you retrain it by each category?”
“You already have the architecture, isn't that half the job?”

In software, when you build an order placement API, it’s reusable, you can lift and shift it across regions.

But in AI, the model trained on one dataset doesn’t behave the same when trained on a subset.
👉 Data imbalances matter. Feature distribution matters. What works in one set may break in another.

2️⃣ Oversimplified expectations

“Can we retrain the model every day?”
“Let’s schedule model updates at the end of each day.”

Nobody trains models every day. That’s not how MLOps, retraining windows, or data quality cycles work.

3️⃣ Confidently asking the wrong questions
The kind of questions we ask reflects our AI awareness rate.
The problem isn’t curiosity, it’s confidence in assumptions without understanding the complexity behind them.

4️⃣ Biases disguised as "opinions"
In many leadership discussions, I observe a mix of:

• Strong opinions shaped by past software patterns
• Lack of exposure to ML trade-offs
• Forcing timelines and expectations AI can't meet,  yet

🔁 This requires unlearning, openness, and re-learning.
AI won't fail because it’s flawed. It will fail when leaders assume how it works — and miss how it actually works.

Let’s not just adopt AI. Let’s understand it.
A little learning, backed by humility, goes a long way.
Titles don’t validate assumptions. Understanding does.

---

#AILiteracy #AILeadership #AIAdoption #MLReality #AIExpectations #EnterpriseAI #TechAwareness #UnlearnToLearn #ResponsibleAI #AIThinking #AIProductLeadership #MLOpsReality #DataMatters

🎯 ML, DL, GenAI - What Do You Really Need?

It’s not about who knows the most models. It’s about who can solve the problem with the right approach.

🚀 In interviews and real-world projects, here’s what separates noise from value:

• Can they choose the right approach? → Classical ML, Deep Learning, or GenAI not everything needs the latest hype.
• Do they know when not to use GenAI? → It’s impressive to know LLMs. It’s smarter to know when not to call them.
• Can they debug when pre-built solutions fail? → You don’t need a model zoo, you need people who can trace the issue and fix it.
• Can they explain their trade-offs and iterate with clarity? → Choosing between latency, accuracy, explainability, and cost is real work.

💡 Skip the overly academic or overly abstract interviews. Hire those who think in problem-first, data-smart, solution-aware ways.

✅ Evaluate with real-world scenarios.
✅ Prioritize learning agility and debugging mindset.
✅ Look for clarity in reasoning, not just complexity in vocabulary.

---

#MLvsDLvsGenAI #AIHiring #GenAIRealityCheck #DataDrivenEngineering #AIProductThinking #ProblemFirst #ResponsibleAI #TechRecruiting #DebuggingMatters #RealWorldAI #InterviewWisdom #EnterpriseAI #ThinkBuildLearn

 Keep Thinking!!!

 

GenAI Reality Check: That RAG Use Case?

 

Keep Thinking!!!

10 Slides in Every GCC Pitch Deck

Adding some Hard truth complementing what it means, for some pages based on observations

10 Slides in Every hashtag#GCC Pitch Deck (and What They Actually Mean)
Slide 4: “We’re Not Just a Support Center”
What it means: But we’re still taking direction from HQ for every single product decision.
✅ Hard truth: Core work may not always be there; the trust factor may not be present.

Slide 5: “India Will Drive Innovation”
What it means: We haven’t defined ‘innovation,’ but it looks great on slide titles.
✅ Hard truth: Many times, it's hackathons that never turn into product features.

Slide 6: “We’ll Scale to 200 People in 12 Months”
What it means: We haven’t built a talent pipeline or local employer brand. But growth sounds impressive.
✅ Hard truth: Sometimes the focus is not on long-term strategy or learning, but on numbers and quantity.

Slide 8: “Hiring Top 1% Engineers”
What it means: We’ll post a JD with 10 must-have skills and wonder why no one’s applying.
✅ Hard truth: Most have every listed skill, follow standard patterns, and are masters of 50 tools, but lack focused depth.

Slide 10: “This Is Just the Beginning”
What it means: It definitely is, and depending on who we hire first, it’ll either scale fast or stall early.
✅ Hard truth: We will provide another narrative when the time changes.

hashtag#perspectives hashtag#observations

Ref - Link

Keep Thinking!!!

From Idea To Prototype With GenAI | Insights by Sivaram A, AI Advisory at Stealth AI

 

Happy Learning!!!

2024 GenAI Tech Talks

Happy Learning!!!

The Great AI Disconnect: What Leaders See vs What Engineers Face

It's a Cycle of Hype - Hope - Repeat

• 🧠 The CEO was sold a “miracle platform” — “It’ll boost your existing models by 90%!”
• 📊 The VP hears, “It integrates 90+ data sources seamlessly!”
• 🧩 The Senior Manager is told, “Plug-and-play pipelines. Just flip a switch!”
• 🧑‍💻 The Engineers? They discover broken schemas, data gaps, and a model that works on benchmarked datasets only

When results don’t match the pitch:

• The CEO is promised: “50% discount next cycle!”
• The VP gets: “Let’s collaborate and fix it.”
• The manager burns out.
• The engineers build workarounds.

And the narrative? Reboots.

🎢 Rinse. Hype. Repeat.

Ref - Link

Keep Thinking!!!