Achintya अचिन्त्य गौमत

@arpit_bhayani It’s not RAG anymore It’s called a Harness now 😂

A lot of production RAG failures are not really LLM failures. In fact, a demo of any RAG system looks rock solid, but when it is shipped to production, things break. Happened with me as well :) So, what does it take to run RAG in production? I wrote a deep dive on what production-grade RAG systems actually require once you move beyond demo and tutorials. It is super practical and filled with code snippets and nuances that need to be taken care of when you are building an agentic RAG system. It covers - RAG basics (for completeness) - how retrieval actually works - why chunking is where most systems fail - embedding model lock-in - reindexing - document registries and chunk identity - index updations and deployments - retrieval, tracing, and observability RAG systems usually fail because of issues with indexing, retrieval, and observability - not because of the model itself. The article focuses on the operational side of RAG systems: keeping indexes fresh and correct over time, avoiding stale retrievals, and building enough tracing to debug bad answers when they happen. If you are building RAG systems on mutable data, give it a read. I kept it pretty practical.

@hellonehha You didn’t get the pun. Her point is americans are not working as hard as indians. Indians easily putting 80-90hrs/week. While americans are putting in 40-50hrs/week. Which is not true, i know

Wrong l. India is outsourcing hub due to low (used to be) cost, more resources , and govt laws. This is a bs take on just coz usa or west countries have wlb , indians are getting job. Do you think an MNC company will take decision to open a new office in a new country , hire 1000 of employees only coz they log out at 4:30PM (lol) Shows your Lack of experience that how private sector or companies work

we eliminated seat-based pricing at @mintlify seat-based saas pricing doesn't have a future where agents are the primary users. it's like pricing electricity by the bulbs

@himanshustwts They both seem to be same. Generalist and synthesist

somehow it only gets more relevant.

@Ken67547214 @yacineMTB Why?

@yacineMTB The word "atlassian" gives me ptsd. Jira is the worst thing that has ever happened to me, professionally.

this video gave me PTSD lmfao

Spent the last two days interviewing candidates for Product Manager roles at a government org. 41 candidates. Startup and big-co PMs. For the first time, I had CVs in front of me showing products I had actually heard of or used. In nine years of doing these interviews, I've never seen this density of talent apply for a contractual government PM role. We used to scrape for under 10 applicants per role, mostly poor fits. I used to banter with friends in big-tech that they hoard good talent and don't really utilize it, while public sector teams and civil society orgs are talent-starved. AI-led restructuring is releasing all that talent now. While we hired really strong candidates for mission-critical roles, I felt very heavy throughout the interviews. It's a bloodbath out there. Stay safe.

@arpit_bhayani I’m just thinking out loud a very basic aspect. LLMs work by just predicting the next token at a time. So if an output 100 tokens (let’s say), so it means 100 iterations of prompt through whole of LLM layers? Is it correct understanding?

I always wondered, when we deal with LLMs, how a temperature parameter brings in variance? Yesterday, I got some time to dig into the internals, and all my learnings around it are compiled in my latest blog. By the way, the blog covers this, the basics, and some math in an easy-to-digest manner. In the essay, I broke down how LLMs actually work at the systems level and explained: - what LLMs are mechanically doing - how next-token prediction really works - what logits and softmax are actually doing internally - why identical prompts can produce different outputs - why hallucinations happen - and how temperature really works under the hood by reshaping the probability distribution before sampling If you are building with LLMs (which I am sure you are), understanding these mechanics gives you a much better mental model. Give it a read, and like always, I hope this helps.

@hellonehha Ok but they are expensive. It will take tech management sometime to realise worth of paying humans more.. in the AI era

In the AI era, you like it or not experienced engineers have lot of EDGE over non-experience engineers. their prompt might not be so sophisticated but they know what is going to break, where it is going to break, how to get it fix or many times just do it by itself. also, other non-functional requirements are where they have the edge.

Underrated advice for job seekers - Reaching out to a founders office instead of an HR will increases your odds of getting a response by a huge margin.

@AgrawalSan38438 @system_monarch No worry. Take your time. Just keep good ai problems coming.

@vibrantachintya @system_monarch Good catch.....that's a bug. The rubric isn't being conditioned on the expected output format, so the grader defaults to expecting code even when the question asks for prose/markdown. Looking into it now, will push a fix. Thanks for flagging 🙏

As an AI Engineer. Please learn: -Prompt caching & semantic caching tradeoffs -KV cache management at scale -Speculative decoding vs quantization -RAG evaluation (RAGAS + human evals) -Cost monitoring & hidden token leaks -Agent guardrails & infinite loop detection

@AgrawalSan38438 @system_monarch Just quick one. First question input is a markdown, but then why it is giving below reasoning about tokens as structure code or something.

@vibrantachintya @system_monarch Glad it came in handy! Try it out and let me know what you think. I'm always open to feedback!

@AgrawalSan38438 @system_monarch This seems pretty helpful. I was looking for something exactly like this.

Thanks for the list! To land an AI Engineering job, I highly recommend practicing on Velocode AI - velocode.ai. It’s a specialized platform with 100+ company-verified problems that actually learn your weak spots as you solve, teaching you the specific concepts needed to bridge the gap.

@vijayshekhar Managing agents of accountability, performance tracking, employee sentiment. Why is everybody obsessed with coding, there are 100s of things to be done in an org. Leaving the coding for the coders 🙌

The thick management layer that manages people will either be managing agent ==> be productive and deliver results or won’t be at all. More output per person employed. Companies will have to do this to survive.

@puneetiitm Indian IT is already earning from labor services. The AI companies are actually targeting to eliminate the labor employed by IT companies

Everyone is reading the Anthropic–Blackstone $1.5B and OpenAI $10B JVs as bad news for Indian IT. It could be the opposite. IT services chased the IT budget — 2–4% of corporate spend. AI services are chasing the labor budget. 10x larger. Demand just got proven. So did the competition to capture it. Blackstone, Goldman, TPG, Brookfield, and Bain don't write large cheques into imaginary profit pools. The AI services cake is real. Much bigger than the cloud. Multi-decadal. The real debate is who eats the slices...

@hellonehha @AnthropicAI Chatgpt workflow diagrams have become so better. I have been using to create dfd on a daily basis.

After reading @AnthropicAI blog on Agentic AI. spent some time to create a mental model to understand how to design, and explain Agentic AI architecture Define a task/goal - what you want agent to do achieve? 1. Orchestration layer : it is your control panel 3. Agents layer: this layers made of agents (multi /specialised) 4. tools: your tools are made of this layer (web search, DB, APIs etc) 5. memory: this is the brain to store information - long or short term etc. 6. monitoring : This is the most crucial to monitor each and every step 7. Reliability & failure management: identify errors, retry, fallback, involve human 8. Governance and security: compliance, audit, auth etc.

@kushalbhagia No need to be offended This is going to be a new norm while hiring/evaluating ideas. I just attempted an AI assessment for a job. Nothing wrong for VC pitches as well. AI is nothing but an extension of humanity, it has collective knowledge of (most) humanity.

As an early stage founder, if a VC firm made you pitch your idea to an AI Agent instead of a human - one that answers back, is generally thoughtful and kind , and even gives quality feedback - would you >feel offended ? > or would take your shot and do your best pitch ?

@HarveenChadha Once a coolie, always a coolie

I don't understand this timeline anymore

Confessions and realities 42M, 55LPA I am a 42-year-old man with a senior job in IT. I have a house in Chennai, a supportive wife, and two children. On paper, everything about my life looks perfect. I have achieved all the things society says a man should achieve. In my twenties, life felt different. I had friends to spend time with. We would hang out at Marina Beach and Besant Nagar beach, watch movies at Rohini, Udayam, and Kasi theatres, and ride around Mount Road on my RX100. In my thirties, I had colleagues to talk with over tea breaks. We would discuss apartments, onsite trips, and share random stories about life and work. But now, in my forties, life has turned into a quiet routine. My phone rarely rings for anything personal. Most calls are about office work, bank alerts, or someone from home asking me to pick up milk on the way back. The loneliness of a man in his forties is unusual. I am not physically alone, but I often feel like a machine. When I enter my home, I am simply “Appa.” I am the person who pays school fees, fixes the Wi-Fi, and handles repairs. My wife is busy with her work and the kids. My children are teenagers now, living in their own worlds and their own rooms. They love me, but they mostly see me as the person who provides comfort and stability. They no longer see me as an individual. At the office, I am the senior person. I am expected to have all the answers. I cannot tell my team that I feel tired. I cannot tell my boss that I sometimes struggle to keep up with new technologies. I must appear confident and strong, even when I quietly worry about the future. Sometimes I drive home slowly from work just to spend a few extra minutes in the car. I listen to songs from my college days. For those fifteen minutes, I am not a manager or a father. I am simply myself again. I realize that I have not had a real conversation about my feelings with anyone in years. My old friends now exist mostly as names on WhatsApp. We send “Happy Birthday” or “Congratulations” messages, but rarely talk. When we meet at weddings, our conversations revolve around our children’s grades or the cars we drive. We never talk about what we actually feel. The hardest part is that I cannot even complain. If I tell my family that I feel lonely, they look confused and say, “But we are all here with you.” They do not understand that a person can be surrounded by people and still feel like they are on a desert island. Society teaches men that if they provide money and security, they have succeeded in life. But no one teaches us how to deal with the silence that comes with it. I have built a beautiful life for everyone around me, but sometimes it feels like there is no space left for me inside it. And maybe… this is what life in your forties feels like.

@thebiryanidev @GanKanchi This is why no one is sad when devs are laid off. Need to show some empathy. You dumbfuck

@GanKanchi 55 LPA is not that high Stop crying and get to work

Something strange is happening in tech. CTOs of billion dollar companies are quitting to take IC roles at Anthropic. Workday CTO -> MTS (Mar 2026) You[.]com CTO -> MTS (Mar 2026) Instagram CTO -> MTS (Jan 2026) Box CTO -> MTS (Dec 2025) Super[.]com CTO -> MTS (July 2025) Adept AI CTO -> MTS (Jan 2025) The mission is that real.