- commitstoday

Ronit Raj

Full-Stack AI & ML Engineer

India · Remote

Full-stack AI & ML engineer who likes shipping - from training and serving models (GRPO on AMD MI300X, vision-language RAG) to the full-stack apps and infra around them. 9x hackathon podiums, a startup, and GitMesh - a Linux Foundation (LFDT) project I maintain.

OSS Mumbai India 2026 Speaker9x Hackathon WinnerGitMesh Maintainer

About

I build because I enjoy solving hard problems. What started with a static HTML website in 2019 has grown into production AI systems — PM-JAY claim adjudication, ClauseMark regulatory evidence, InferenceGym for LLM serving, and PARAM on-chain credentials. Along the way, I've earned 9 hackathon podiums ($5,000+ in prizes), co-founded a startup, and maintain GitMesh, an LFDT AI-agent governance project at 140+ stars. Today, I focus on full-stack AI and AI infrastructure, turning prototypes into systems people can rely on.

Where It Started

In 2019 I shipped my first website - a static hotel landing page with responsive layout, a mobile side nav, and vanilla JavaScript (still on GitHub as first-website). No framework, no backend - just HTML, CSS, and curiosity. That foundation is the through-line to today: PM-JAY pipelines on Gemma, ClauseMark for UN digital trade, PARAM on Base Sepolia, InferenceGym for LLM serving, and a 40-minute session at OSS India 2026.

My Journey in AI

I started with a simple question at IIIT Gwalior: how do you make complex systems actually work? That curiosity grew into building production AI - from military intelligence platforms to on-chain applications. Winning Smart India Hackathon 2024 with a 79%-accuracy RAG system taught me the lesson that shapes everything I build: great AI isn't fancy architectures, it's shipping systems that hold up in production.

The Infra Underneath

The part I keep coming back to is the infrastructure beneath the model. I built InferenceGym, an RL environment that treats LLM serving as a control problem - batching, KV-cache budgets, speculative decoding, prefill/decode splits - and trained a policy that beats hand-tuned heuristics. I've fine-tuned with GRPO and LoRA on AMD MI300X, deployed models as Docker Spaces on Hugging Face, and built RAG pipelines over Qdrant and FAISS. I've shipped on-chain systems too - academic credentials, a micropayment CLI - when decentralization actually earns its place. But what pulls me in is making inference fast, cheap, and reliable at scale.

From Hackathons to Production

Build fast, ship faster, iterate. Co-founding Garudex Labs taught me the gap between demo and production is where engineering happens. Whether shipping EdTech serving 50+ students or maintaining open-source tools like GitMesh, the real work is error handling, feedback loops, and making sure the system doesn't break when real users hit it.

Technical Approach

PyTorch and Transformers for the models; FastAPI and Next.js for delivery; whatever serving stack keeps latency and cost sane. I reach for smart contracts only when decentralization adds real value. Mostly I optimize for one thing: code other engineers can read, deploy, and extend - because the best system is the one that ships and keeps working when I'm not watching it.

Work Experience

GitMesh

Open Source

Core Maintainer

Core Maintainer of GitMesh — a Linux Foundation multi-agent orchestration runtime with OPA policy-as-code and MCP governance. Grew adoption to 140+ GitHub stars.

July 2025 - PresentRemote

C3iHub, IIT Kanpur

Internship

Software Engineer Intern

Directed PARAM — a production blockchain academic-credential platform with a live pilot across 6 departments and 6 degree programs at ABV-IIITM Gwalior.

May 2025 - August 2025Remote

Ionia

Full-Stack Developer

End-to-end Online Testing Infrastructure with pilot 50+ active users

December 2024 - June 2025Remote

Education

Indian Institute of Information Technology and Management, Gwalior

B.Tech in Mathematics and Scientific Computing

Specializing in AI/ML and Data Science

2023 - 2027CGPA: 8.35

Featured Projects

Production AI, full-stack apps, and on-chain systems - with demos and case studies.

View All →

Linux Foundation · LFDT

GitMesh

Full-Stack AIAI Infrastructure

Policy-as-Code Engine for Open-Source AI Agent Orchestration

Stack

TypeScriptNode.jsOPA / RegoMCP+6

Demo GitHub Case Study

NHA × IISc · Top 3 Runner-Up · $3,100

PM-JAY Claim Adjudication

MLHealthcare/AI

Automated Evidence Verification for Ayushman Bharat

Stack

Gemma 3 12BPyMuPDFOpenCVPillow+5

GitHub Case Study

UN Digital Trade Hackathon · 2026

ClauseMark

MLLegalTech/AI

Verified Regulatory Evidence Engine - UN Digital Trade (RDTII)

Stack

PythonLangGraphFastAPIQdrant+8

GitHub Case Study

C3iHub · ABV-IIITM Gwalior

PARAM

Web3EdTech/Blockchain

Blockchain-Based Academic Credential Verification for ABV-IIITM Gwalior

Stack

Next.js 16React 19TypeScriptTailwind+13

Demo GitHub Case Study

SIH 2024 Winner · $300

SHAKTI

MLAI/ML

Vision-Language Model (SHAKTI-2B) for Military Intelligence

Stack

mPLUG-Owl3MS-SWIFTRAGFastAPI+2

Demo GitHub Case Study

Meta × PyTorch × Hugging Face Hackathon

InferenceGym

MLAI/ML

OpenEnv RL Environment for LLM Inference Serving

Stack

PythonPyTorchPPOOpenEnv+4

Demo GitHub Case Study

Achievements

Speaking

Hackathon Wins

9×$5,000+

Patent

More Achievements

Skills

◇PyTorch◇Transformers◇RAG◇Reinforcement Learning◇GRPO◇PPO◇LoRA◇PEFT◇Unsloth◇TRL◇Bitsandbytes◇LangChain◇LangGraph◇Graph Neural Networks◇Recommender Systems◇Hugging Face◇vLLM◇Docker◇Kubernetes◇ROCm◇MCP◇Weights & Biases◇GitHub Actions◇NumPy◇Pandas◇Pydantic◇FastAPI◇Flask◇OpenCV◇BeautifulSoup◇Selenium◇Next.js◇React.js◇Node.js◇Express◇GraphQL◇WebSocket◇React Native◇Tailwind◇PostgreSQL◇MongoDB◇Redis◇Supabase◇Prisma◇Drizzle◇Neo4j◇SQLite◇Pinecone◇FAISS◇Qdrant◇ChromaDB◇Solidity◇ethers.js◇Foundry◇Base◇AWS◇GCP◇Vercel◇Cloudflare◇Python◇TypeScript◇JavaScript◇SQL◇C/C++◇HTML◇CSS

Showing 66 total skills