Pratyush Choudhury (PC)
@177pc
Enabling more AI in India | 💫 x: @awscloud | Views my own | DMs: http://superdm.me/177pc
I like @deedydas's work but but this take misses context Sarvam-M isn’t a vanity fine-tune; it’s India’s first open-weights 24 B Indic-centric LLM built under brutal GPU & data scarcity. Judging it by few hours of HuggingFace stats badly misses the point. Most people outside…
India's biggest AI startup, $1B Sarvam, just launched its flagship LLM. It's a 24B Mistral small post trained on Indic data with a mere 23 downloads 2 days after launch. In contrast, 2 Korean college trained an open-source model that did ~200k last month. Embarrassing.
Agents shouldn’t break like cheap pens 🖊️ Pumped to see @composiohq bag $25M Series A led by friends at @lightspeedvp + a dream bench of operators to ship the "skill layer" where agents learn-on-the-job The current traction numbers are 🔥- With 100k devs, 10M+ daily calls, it…
Agents aren’t reliable. They don’t learn from experience. At @composiohq, we provide skills that evolve with your agents @lightspeedvp gave us $25M to make agents usable
I've followed @sachdh's love for RL for 2+ years at this point & it's so great to see him help ship this This is interesting for a whole host of reasons but in particular because "scope" is beating "scale" Enterprise AI is tilting from “giant‑generalist” AGI aspirations toward…
Excited to share Aryabhatta 1.0, our leading model that scores 90.2% on JEE Mains, outperforming frontier models like o4 mini and Gemini Flash 2.5 Trained by us at @AthenaAgentRL , in collaboration with @physics__wallah, using custom RLVR training on 130K+ curated JEE problems…
An honour to host @vishalmisra (Vice Dean of AI @CUSEAS & the accidental inventor of RAG) for an exclusive session w/ Bangalore's top AI builders. Vishal was predicting GPT's capabilities & limitations much before anyone else. We skipped the hype & dove deep into what's next🧵…
🗓️ AI Pulse⚡️| June 28–July 6: Power, Talent, Paywalls, Services & more in AI in the past week in India &🌏 🎬 TL;DR: This week was about securing power, distribution rights to the next interface & access rights to the best data sets + talent to build the next set of AI…
Solid work out of @lossfunk @encapsulated007 Treating GPU as an evolution arena & moving away from "write-once, run-many" to "spawn-many, distil-one" could be an interesting idea for further work Sharing some thoughts outside-in that might be useful to tinker more with: 1/ GPU…
New blog post: We've never enjoyed working on Kernels more than this. We have some very fast AI-generated kernels with a simple multi-agent system. They're running close to or even surpassing Pytorch shipped kernels. (1/6) [🔗 link in final post]
Good call by @emollick @mathemagic1an From my vantage point, Voice AI is shifting from novelty (smart speakers, "fake friends") to mission-critical agentic workflows wherever hands-busy, high-stakes or multi-language interactions block productivity or revenue. Three seams are…
You have access to an API that can have a pretty great conversation with humans that people actually seem to enjoy, as well as the ability to access tools, websearch, and run code. Somehow, founders seem to have trouble thinking of applications beyond therapist or fake friend...
🗓️ 3rd edition of⚡️AI Pulse (21st-27th June '25): This week, AI’s scoreboard shifted - from parameter counts to watts, wafers & rulebooks 🇮🇳🇺🇸 1/ Infra is increasingly becoming the new edge: - White House drafts executive orders to unlock grid access & federal land for AI…
RL-fine-tuning (RLFT) & declarative self-improvement (DSI) have crossed the “works-out-of-the-box” threshold. Off-the-shelf PPO/GRPO variants, abundant synthetic reward signals & turnkey infra (OpenAI Responses API + Agents SDK, @DSPyOSS) now deliver upto +20-60 pp task lifts…
I wouldn't have said this 6 months ago, but I now believe all serious agents will be RL'd on their specific task. The gains are too easy and too huge to ignore. Either @OpenAI et. al. will provide APIs to do this on-platform, or open source will win.
.@karpathy's framing crystallizes something fundamental here Context engineering represents a paradigm shift from deterministic prompt optimization to probabilistic context orchestration - fundamentally changing how we architect AI systems. Industrial-grade LLM/AI applications…
+1 for "context engineering" over "prompt engineering". People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window…
🗓️ 2nd edition of⚡️AI Pulse: The past week reset the pieces on the global AI chessboard & had some interesting implications for AI in India🇮🇳 Executive Summary: Meta's ~$14.5B non-voting, 49% bet on Scale AI telegraphs that curated data + annotation talent are now the hard…
This was a solid listen from @latentspacepod w/ severe; key ideas, @polynoamial emphasized that the recent advances from OpenAI (the o-series reasoning models, Deep Research, Reinforcement Fine-Tuning) signal a decisive shift from training-time scaling to test-time compute (TTC):…
There's a new wave of GEO-over-SEO tooling companies, @tryprofound being the most notable alongside @withdaydreamco that are reframing where & how brands fight for visibility. They shift budgets from traditional keyword-rank dashboards toward “model-mention” analytics &…
IN NEWS: @tryprofound raises a $20M Series A led by @kleinerperkins. "The tailwinds are mind-blowing. This is a clear platform shift, maybe one of the biggest in the history of marketing." - @thejamescad "How your brand shows up in these AI responses is becoming boardroom-level…