Devendra singh shekhawat

Devendra singh shekhawatJun 3, 2026

Too many vector databases. Everyone is confused about which one to pick. Let me make it dead simple with real examples. Think of a vector database like a smart library. Normal database = finds the book with exact title you type. Vector database = finds books with similar ideas, even if the title is different. Now which library do you build? Depends on your situation. 🧪 Just starting out or building a demo? Use Chroma. Example: You're a student building a chatbot that answers questions from your college notes. Chroma runs on your laptop, zero cost, zero setup. Perfect. 🚀 Launching a real product but small team, no DevOps? Use Pinecone. Example: A startup building an AI customer support bot. They don't have engineers to manage servers. Pinecone is fully managed — just plug in and focus on the product. ⚡ Need performance and want to save money at scale? Use Qdrant. Example: A mid-size company with 10 million product embeddings for their search engine. Self-hosting Qdrant costs ₹2,500/month. Same thing on Pinecone? ₹25,000+/month. Same result. 10x cheaper. 🔍 Need AI search + keyword search together? Use Weaviate. Example: A legal tech company building a document search tool. Lawyers search by exact case names AND by meaning. Weaviate handles both natively — no extra setup. 📦 Already using Postgres for your app? Use pgvector. Example: A SaaS app that already stores user data in Postgres. Just add pgvector as a plugin. No new database. No new server. Done. 🏢 Billions of records? Enterprise scale? Use Milvus. Example: An ecommerce giant like Flipkart running recommendation engines across 500 million products. Only Milvus handles that kind of scale comfortably. The real advice? Don't pick based on hype. Pick based on your team size, budget, and scale. Start with Chroma. Graduate when you need to. Which one are you using? 👇 #VectorDatabase #RAG #AIEngineering #TechIndia #LLM #MLOps

Devendra singh shekhawatJun 8, 2026

We just launched the new adrsta.ai — and honestly I still can't believe how fast it happened. We had an older version running. It worked. But after a team meeting we all felt the same thing — it's missing a lot, messaging is off, it doesn't feel like us anymore. So we decided to redesign. We spent time researching. Looked at what good websites in our space look like, what sections work, what doesn't. Combined all of that with our own ideas and what we already had. Then I just decided — let's use AI and ship this fast instead of going back and forth for weeks. And this is where it got interesting. I literally just sat and started talking through what I wanted. Section by section. Page by page. My thoughts, the references we had, the feedback from meetings — all of it went in as context. And it just... built. I reviewed, took it to the team, got feedback, came back with more context, iterated. But honestly the biggest thing I took away: It's not about how good you are with AI. It's about how clear you are with yourself. If you know what you want to build, you have a direction, you're intentional with your context — AI will execute without losing focus. The moment YOU are confused, the output gets confused too. Best practices and clarity of thought matter more than ever. AI just collapses the time between idea and execution. That part genuinely felt like magic. Would love to hear what you think. 👇 https://www.adrsta.ai @Adrsta #Adrsta #BuildInPublic #ShipFast #MarketingScience #AITools

Devendra singh shekhawatJun 1, 2026

One of the most confusing decisions in AI right now — RAG or Fine-tuning? Let me just break this down simply. RAG is like giving AI an open book exam. You don't change the AI at all. You just keep updating the book. Fine-tuning is like making the AI study for months. The knowledge gets permanently baked in. No book needed. When I'd go with RAG: Your data keeps changing. Like a legal or compliance tool where rules update every month. You can't retrain a model that often. Just update the docs. You need the AI to say "source: XYZ policy, updated April 2025." RAG can do this. Fine-tuning can't. You're early stage and watching costs. RAG is way cheaper to build and run. When I'd go with Fine-tuning: You want the AI to always sound a specific way. Like a bank that needs every response to be formal and on-brand. Fine-tuning changes how the model speaks. RAG doesn't. Your data is too sensitive to sit in a retrieval database. Government or defence use cases. Bake it into the model instead. You need speed. RAG adds latency because it has to fetch docs first. Fine-tuned model just answers from memory. Honestly? Start with RAG. Almost always. It's faster, cheaper, and easier to fix if something goes wrong. Fine-tune only when you've hit a clear wall with RAG. What's your experience — RAG or fine-tuning? 👇 #MLOps #RAG #LLM #AIEngineering #TechIndia