harshit = {
"role" : "AI Safety Engineer | Applied AI Systems",
"focus" : ["Agentic AI", "Multi-agent Orchestration", "AI Evaluation", "On-device ML"],
"building" : ["AIOS"],
"researching": ["PTF โ Agentic AI Failure Study (SSRN)", "Genesis Lock โ Blockchain AI Governance"],
"philosophy" : "From math โ production. No API wrappers, no shortcuts.",
"location" : "India ๐ฎ๐ณ โ India-first deployment always.",
}|
Empirical study of an agentic AI failure incident An investigation into a real multi-agent system failure โ root cause analysis, failure cascade mapping, and safety mitigations. Submitted to SSRN. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6472278
|
| Project | What | Status |
|---|
| ๐ฅ๏ธ AIOS | Local offline AI desktop assistant + AI-Native Operating Environment โ zero cloud dependency | โ Live |
| Project | Signal | |
|---|---|---|
| โก | Anthropic Performance Challenge | 10.2ร kernel speedup on Anthropic's real hiring take-home. VLIW cycles: 147K โ 14K |
| ๐ง | micrograd-JAX | Rebuilt Karpathy's autograd in JAX โ grad, jit, vmap, XLA-compiled, GPU-ready |
| ๐ฅ๏ธ | AIOS | Local AI desktop assistant. Ollama-powered. Built before on-device AI was a trend |
| ๐ | ML Fundamentals | 19 notebooks โ every core ML algorithm from scratch in NumPy. #100DaysOfML |
| ๐ญ | CodeAlpha ML Internship | Production ML: credit scoring, image recognition, disease prediction, speech emotion |
I'm actively looking for ML Engineer / Applied AI / AI Safety / AI Systems roles โ remote globally or India-based.
If you're building in agentic AI, multilingual NLP, on-device inference, or AI safety โ let's talk.



