Hi, I’m Pranay – I build open-source AI that works offline, learns fast, and gives power back to devs.
AI should be accessible, private, and fast – not hidden behind APIs.
That's why I build tools like:
- Yudai v2 — a self-hosted, prompt-based data analyst
- solo-server — a one-command LLM playground for Qwen, DeepSeek, and more
- DeepSeek-R1 Distillation — distilled a 7B reasoning model to 40% SWE-bench pass@1 on a single GPU
My goal? Empower indie hackers, researchers, and builders with local-first AI stacks.
- Llama Impact Grant Winner – recognized for pushing open-source AI tooling ([announcemet link - https://x.com/pranay5255/status/1917873008758456630))
- solo-server OSS maintainer – 300+ indie devs using it to deploy local models in seconds
- Yudai v2 – offline AI analyst for product teams (self-hosted)
- National-level hackathon mentor – mentored 50+ teams; winners at Smart India Hackathon, Prayatna 2.0 at AITR university.
- Deep Web3 Infra Contributor – built protocol tools (AI explainer bots, AI agents for defi, Twitter bots for yapping) for Mode, FortyTwo money
- Top 50 Global Kernel Founders (KB8) – selected into Gitcoin’s elite founder cohort driving innovation in AI x Web3
- Finalist, MEGAZU Pop-up City – chosen among top engineers globally to build Web3 infra with EigenLayer, Ethereum foundation and MegaETH.
- Petabyte-scale ETL @ CoinSwitch – production Spark/Airflow pipelines for ML + risk systems
- Vgyaan (pre-GPT) – BERT-powered edtech system that answered 120k+ questions/night
- Core AI builder at heart – I live for shipping 👷♂️→🚀
📜 Résumé Snapshot
Senior ML / GenAI Engineer • 8 yrs in AI, 2 yrs in crypto infra Domains: LLMs, generative agents, on-chain AI, distributed data systems Highlights: solo-server maintainer, Llama Grant winner, Kernel Founder, Web3 finalist @ MEGAETH Mission: Build tools that give people superpowers, not cloud lock-in.
These shape how I think about training, distilling, and deploying performant LLMs:
- DeepSeek-R1 – reasoning via RL and BoN distillation
- LoRA Insights by Lightning AI – parameter-efficient fine-tuning on low-resource GPUs
- Reinforcement Learning with Verifiable Rewards (RLVR) – reward guarantees for reasoning tasks
- TabPFN – foundation models for small-data tabular learning
- Absolute Zero Reasoner (AZR) – few-shot zero-reasoning for complex reasoning in LMs
- GRPO: Group-based Reinforcement Policy Optimization – advances over PPO, novel RL algorithm from DeepSeek
- NoFeeSwap Yellow Paper – zero-spread market making, liquidity growth AMM design
- SWE-smith: Scaling Data for Software Engineering Agents – open-source pipeline for scalable code bug synthesis, dataset & agent finetuning for SWE-bench
- debug-gym: A Text-Based Environment for Interactive Debugging – interactive, tool-augmented agent environment for LLM-based debugging
- 🔭 Current Projects: - Yudai v2 – offline data analyst for product & growth teams - solo-server – localhost drop-in server for open LLMs
- 🧠 Learning: Verifiable rewards, activation tricks
- ❤️ Philosophy: I thrive on exploring AI and crypto research, and I'm passionate about shipping products that users genuinely find valuable and love to use.
