Neelam Yadav

A practical GenAI lesson: Do not trust AI because it explains itself well. Trust it because you built systems around it that make it reliable. For me, that means: ➡️ retrieval ➡️ grounding ➡️ policy checks ➡️ evals ➡️ human verification where needed AI trust is a system design problem.

A practical GenAI lesson: Do not trust AI because it explains itself well. Trust it because you built systems around it that make it reliable. For me, that means: ➡️ retrieval ➡️ grounding ➡️ policy checks ➡️ evals ➡️ human verification where needed AI trust is a system design problem.

A simple way to think about hallucination: Sometimes AI does not hallucinate because it wants to make things up. It hallucinates because it thinks it knows the answer when it actually doesn’t. That’s a huge difference. The fix is not just “prompt better.” The fix is better grounding, retrieval, and validation.

𝗧𝗵𝗲 𝘄𝗼𝗿𝗹𝗱 𝗻𝗲𝗲𝗱𝘀 𝗺𝗼𝗿𝗲 𝗘𝘂𝗴𝗲𝗻𝗲 𝗥𝗼𝘀𝗵𝗮𝗹. The 𝗺𝗮𝗻 𝗯𝗲𝗵𝗶𝗻𝗱 𝗪𝗶𝗻𝗥𝗔𝗥. And yes, RAR literally stands for Roshal Archive. Back in 1993, he created the RAR format. Then in 1995, he released WinRAR for Windows. And then he accidentally created one of the greatest distribution hacks in software history: “𝟰𝟬-𝗱𝗮𝘆 𝗳𝗿𝗲𝗲 𝘁𝗿𝗶𝗮𝗹.” But here’s the funny part everyone remembers: The reminder showed up. People clicked OK. And kept using it anyway. For years. Sometimes decades. No drama. No aggressive lockouts. No “your files are blocked until you pay” nonsense. Just a product that solved a real problem: smaller files easier sharing cheaper storage better compression That’s why WinRAR spread everywhere. School computers. Office desktops. Internet cafes. Home PCs. In a world obsessed with growth hacks, WinRAR won with something simpler: make useful software and become impossible to uninstall from human habit. Absolute legend. #WinRAR #EugeneRoshal #Software #TechHistory #BuildInPublic

𝗚𝗼𝗼𝗴𝗹𝗲 𝗱𝗶𝗱𝗻’𝘁 𝗷𝘂𝘀𝘁 𝗹𝗮𝘂𝗻𝗰𝗵 𝗚𝗲𝗺𝗺𝗮 𝟰. They launched a 𝗟𝗢𝗖𝗔𝗟 𝗔𝗜 𝗳𝗹𝗲𝘅. 🔥 Everyone talks about benchmarks. I care about this: Can I run it on my own machine? With Gemma 4, the answer is yes. Google’s official docs explicitly support local/on-device usage, and publish memory guidance for local inference too. The standout: 26B A4B MoE 26B total params 4B active per token So this launch is not just about model size. It’s about making serious AI more practical locally. #Gemma4 #GoogleAI #MoE #OpenSourceAI #LLM #GenAI

One thing I’ve noticed with AI: Sometimes it gives the right answer but the explanation it gives for how it got there is not the real one. That’s why in production, I don’t trust “nice reasoning steps” blindly. Lesson: Good output does not always mean faithful reasoning. Always verify

@NehraWorkss Lets connect

If you are in tech say hi 👋🏻 Let's connect

Its the reality. Every non tech person has now become highly technical person in there opinion

@waelnassaf Lets connect. I am into building AI tools

I'm cleaning up my X followings list: - No news agencies - No fake influencers - No negative noise + Just creators and founders looking to make it big in 2026 If you are one, let's connect 🤝

@SurajJ326311 Lets connect

Hi everyone 👋 I’m a Software Engineer passionate about building things and exploring new tech. If you're in the tech space, let’s connect and learn from each other 🤝

@ganesh__47 I am building AI tools

@XTechOps Hey ! Happy to connect! What are you building?

Hi guys 👋 I’m a Software Engineer passionate about building and exploring new tech. If you’re in tech, let’s connect 🤝

RAG vs Agentic RAG : Lets simplify it RAG = User asks a question → system retrieves relevant documents → LLM answers from that context 📚 It is great for: ✅ grounded answers ✅ enterprise knowledge search ✅ reducing hallucinations ✅ fast Q&A over your data But mostly, it is still a retrieve + respond pattern. Agentic RAG = The system does not just retrieve once and answer. It can: 🔎 rewrite the query 🧭 decide which source/tool to use 📂 retrieve from multiple places 🧪 validate what it found 🔁 loop again if context is weak 📝 then generate the final answer So the shift is: RAG = fetch context and answer Agentic RAG = reason, plan, retrieve, verify, then answer #AI #RAG #AgenticAI #AgenticRAG #LLMOps #GenerativeAI #EnterpriseAI

𝗖𝗼𝗺𝗽𝗼𝘀𝗲𝗿 𝟮 𝗶𝘀 𝗻𝗼𝘄 𝗮𝘃𝗮𝗶𝗹𝗮𝗯𝗹𝗲 𝗶𝗻 𝗖𝘂𝗿𝘀𝗼𝗿. This feels like another strong signal in the evolution of AI coding. Not just faster autocomplete, but more capable agentic coding inside the developer workflow. 𝗪𝗵𝗮𝘁 𝘀𝘁𝗮𝗻𝗱𝘀 𝗼𝘂𝘁: ⚡ frontier-level coding performance 💸 strong intelligence-to-cost balance 🧠 Cursor’s own model for agentic coding 🚀 available with both standard and fast variants Cursor says Composer 2 is now live in the product, with pricing starting at $0.50/M input and $2.50/M output tokens, plus a faster default variant. We are moving from: 𝘈𝘐 𝘵𝘩𝘢𝘵 𝘩𝘦𝘭𝘱𝘴 𝘸𝘳𝘪𝘵𝘦 𝘤𝘰𝘥𝘦 to 𝘈𝘐 𝘵𝘩𝘢𝘵 𝘤𝘢𝘯 𝘱𝘢𝘳𝘵𝘪𝘤𝘪𝘱𝘢𝘵𝘦 𝘮𝘰𝘳𝘦 𝘢𝘤𝘵𝘪𝘷𝘦𝘭𝘺 𝘪𝘯 𝘴𝘰𝘧𝘵𝘸𝘢𝘳𝘦 𝘦𝘯𝘨𝘪𝘯𝘦𝘦𝘳𝘪𝘯𝘨 𝘸𝘰𝘳𝘬𝘧𝘭𝘰𝘸𝘴. Interesting times ahead for developers, platform teams, and anyone building with AI-native tooling. #Cursor #Composer2 #AICoding #GenAI #DeveloperTools #SoftwareEngineering #AgenticAI

Where do you get more value today - X or LinkedIn? 👀

If you are looking for interview calls as a GenAI Developer, don’t just list skills. Add: ✅ 1 solid RAG project ✅ 1 real end-to-end GenAI project That alone can increase your chances of getting shortlisted. Try it. Thank me later. #GenAI #RAG #AgenticAI #LLMOps #AIJobs

@KevinBelfort_30 lets connect

Builders on X, What are you working on right now? • SaaS • AI tools • Startup * • Side project Drop it below Let’s connect 🤝

@impeculiar1b lets connect

If you are still under 5k followers Say hi 👋 Let’s follow you

@omjain23 lets connect

twitter is cool. but it’s 100x better when your timeline is full with people who code and build things. if you’re into tech, AI, startups, design, web dev, web3, or programming, say hi 👋

🔥 Below are some of the questions orgs are asking GenAI Developers these days to test whether they have actually built real systems: 🔹 Walk me through a GenAI app you built end to end 🔹 If your RAG system is hallucinating, how would you debug it? 🔹 How did you choose chunking strategy and what broke when it was wrong? 🔹 Have you ever changed the embedding model? How did you migrate safely? 🔹 What evals did you build before release? 🔹 How do you know whether the issue is in retrieval, prompt, or model? 🔹 Tell me about a real production incident in your GenAI system 🔹 When did you choose workflow over agents and why? 🔹 How do you prevent loops, bad tool calls and cost blowups in agents? 🔹 How did you add citations, trust signals or groundedness in RAG output? 🔹 What did you do to reduce latency and cost in production? 🔹 What logs, traces and dashboards did you monitor? 🔹 How did you handle tenant isolation / ACL in RAG? 🔹 How did you protect against prompt injection or unsafe tool use? 🔹 If I give you a broken RAG pipeline today, what 5 things would you inspect first? This is where interviews are moving: from “Do you know GenAI?” to “Have you actually shipped it?” #GenAI #RAG #AgenticAI #LLMOps #AIJobs