██████╗ ██████╗ ██╗ ██╗██╗████████╗ ██╗ ██╗ █████╗ ██████╗ █████╗ ██╗ ██╗ ██╔══██╗██╔═══██╗██║ ██║██║╚══██╔══╝ ╚██╗ ██╔╝██╔══██╗██╔══██╗██╔══██╗██║ ██║ ██████╔╝██║ ██║███████║██║ ██║ ╚████╔╝ ███████║██║ ██║███████║██║ ██║ ██╔══██╗██║ ██║██╔══██║██║ ██║ ╚██╔╝ ██╔══██║██║ ██║██╔══██║╚██╗ ██╔╝ ██║ ██║╚██████╔╝██║ ██║██║ ██║ ██║ ██║ ██║██████╔╝██║ ██║ ╚████╔╝ ╚═╝ ╚═╝ ╚═════╝ ╚═╝ ╚═╝╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝╚═════╝ ╚═╝ ╚═╝ ╚═══╝
🖥️ About • 💼 Experience • 🚀 Projects • 🛠️ Stack • 📄 Resume
I'm Rohit Yadav, a 4th-year Information Technology student at NIT Jalandhar , and I don't just use tools, I build them.
I specialize in building systems end-to-end , from low-level infrastructure (bootloaders, virtual machines, compilers, operating systems) to AI/ML, genomics, and large-scale data pipelines.
- 🧠 Built 25+ LLMs from scratch , 120M to 1.4B parameters on TPU clusters
- 🔬 Trained on 500 Billion+ tokens using Google's TPU Research Cloud
- 🏗️ Built a complete Compiler → Virtual Machine → OS pipeline from scratch
- 🧬 Created a 500M-param DNA Foundation Model outperforming AlphaMissense & CADD
- 💡 No frameworks, just raw memory and metal
| Role | Organization | Duration |
|---|---|---|
| 🤖 AI Research Intern | Molsys Pvt. Ltd. | Jun 2026 – Present |
| 🔬 Research Lead | AutonomousX | Apr 2026 – Present |
| 🎓 ML Summer School | Amazon (Selected: 3,000 / 134,000+ applicants) | Jun 2026 – Jul 2026 |
| 🌐 Open Source Contributor | Google DeepMind Gemma · Meta LLaMA | Mar 2026 – Present |
| 🖥️ Full-Stack Web Dev Intern | Skylark Express Delhi Pvt Ltd (On-site, Gurugram) | Jun 2025 – Jul 2025 |
AI/ML·LLMs·Transformers·Distributed Systems·JAX·TPU
- 🏗️ Built and open-sourced 25+ LLMs (120M–1.4B params) by engineering TPU-native distributed pipelines (JAX pmap) under AutonomousX
- 📊 Trained the Instinct model family on 500B+ tokens (PILE, DOLMA) under extreme overtraining regimes to study scaling laws, convergence, and degeneration dynamics
- 📹 Open-sourced complete pipelines including HuggingFace weights, training logs, and a 10-minute TPU setup guide
- 🧪 Provided a ready-to-run inference notebook for quick evaluation
██████╗ █████╗ ██████╗ ███████╗ ██████╗ ██████╗ ██████╗ ███╗ ██╗ ██╔══██╗██╔══██╗██╔════╝ ██╔════╝ ██╔══██╗██╔═══██╗██╔══██╗████╗ ██║ ██████╔╝███████║██║ ███╗█████╗ ██████╔╝██║ ██║██████╔╝██╔██╗ ██║ ██╔══██╗██╔══██║██║ ██║██╔══╝ ██╔══██╗██║ ██║██╔══██╗██║╚██╗██║ ██║ ██║██║ ██║╚██████╔╝███████╗ ██████╔╝╚██████╔╝██║ ██║██║ ╚████║ ╚═╝ ╚═╝╚═╝ ╚═╝ ╚═════╝ ╚══════╝ ╚═════╝ ╚═════╝ ╚═╝ ╚═╝╚═╝ ╚═══╝
AI Agents·LLM Systems·Terminal Automation·Browser Agents
- 🤖 Built an AI-powered terminal agent for Linux with 25+ integrated tools enabling autonomous terminal execution, file editing, web search, browser automation, and Google Workspace operations through natural language
- 🧠 Designed a planner → tool execution → feedback loop with persistent memory, multi-terminal sessions, background task management, and safety confirmation for destructive commands
- 🌐 Implemented Playwright-based browser automation, persistent profiles, Google Workspace integration, and intelligent web research
- 🔧 Engineered a production-style CLI with modular tool architecture, task history, markdown rendering, and extensible plugin design
AI/ML·GenAI for Biology·Foundation Models·Bioinformatics
- 🧬 Developed a hybrid genomic foundation model (500M params) fusing a Nucleotide Transformer backbone with clinical tabular features for SNV pathogenicity prediction
- 📈 Achieved 0.9186 ROC-AUC , outperforming industry SOTA (AlphaMissense, CADD) by +0.32 ROC-AUC
- 🏥 Designed a ranking-first clinical triage system with 75%+ pathogen recall while testing only 5–10% of variants
- 🔍 Validated via SHAP ablation studies (64.9% reliance on raw DNA) and open-sourced the full pipeline
letbro a = 10;
letbro b = 3;
printbro(a + b);
letbro a = 10;
letbro b = 3;
printbro(a / b);
letbro a = 10;
letbro b = 3;
printbro(a - b);
letbro a = 10;
letbro b = 3;
printbro(a * b);
ifbro (a > b) {
printbro(999);
} elsebro {
printbro(111);
}
ifbro (a < b) {
printbro(222);
} elsebro {
printbro(888);
}
ifbro (a == b) {
printbro(333);
} elsebro {
printbro(777);
}
letbro counter = 0;
whilebro (counter < 3) {
printbro(counter);
letbro counter = counter + 1;
}
Systems Programming·Compilers·Runtime Systems·C++
- 🔧 Designed a complete compiler from scratch for a custom language , implementing Lexer, Parser, AST, and Bytecode Generator
- 💻 Built a custom 16-bit Virtual Machine (VCPU + 65KB RAM) , inspired by Java + JVM-style execution
- ⚙️ Full runtime system: instruction decoding, stack-based execution, control flow, and I/O operations
- 🛠️ Tech: Modern C++17, custom ISA design, bytecode architecture, manual memory management
| ☁️ | Google TRC Grant , 320 TPU access (SPOT + STANDARD VMs) |
| 🔴 | AMD MI300X GPU , 300 non-preemptive hours |
| 🎓 | Amazon ML Summer School 2026 , Selected among 3,000 from 134,000+ applicants |
| 🧮 | IOQM (PRMO) , Qualified for Indian Olympiad Qualifier in Mathematics [Certificate] |
| 🔬 | KVPY SA , AIR 2590 [Certificate] |
| ⚔️ | LeetCode , Knight (2013), 600+ problems [Profile] |
| ⭐ | CodeChef , 4-Star (1818) [Profile] |
| 🏅 | Codeforces , Specialist (1417) [Profile] |
| 🥇 | CodeChef START166B , Global Rank 9 [Contest] |
I'm always open to interesting conversations, collaborations, and opportunities!
|
yrohit1825@gmail.com |
Rohit Yadav |
YADAV1825 |
AutonomousX |
|
rohit-portfolio |
tomodachi_ |
tomodachi_ |








