Computer science student building agentic systems, applied ML, and full-stack tools — and shipping them end to end instead of leaving them in a notebook.
I'm a fourth-year CS student at Jain University, mostly working at the intersection of LLM agents, applied ML, and backend systems. I'd rather understand how something works than just use it — that's what pulled me toward AI in the first place, and it's also why most of what's below is a full working system rather than a notebook.
Recently: co-authored an accepted paper on prompt optimization, and submitted an Agent Skill to the official marketplace (pending review). Currently building an online cubing platform on the side, and working through the JLPT.
| Project | What it is |
|---|---|
| TraceFlow | Multi-agent root-cause-analysis engine — three agents gather evidence, write, and verify an RCA report for a factory fault, behind real auth, tenant-scoped data, and an eval harness. |
| Drifty | Prompt refinement and drift monitoring — three models draft and rank candidate prompts, then a hand-rolled Redis queue re-runs the winner daily and fires a webhook the moment it starts failing. |
| Context Handoff | An Agent Skill that keeps two living documents — architecture and task log — so a new AI chat session can pick up a project with zero context loss. |
| Explainable Eye Disease Detection | CNN vs. Swin Transformer on retinal images, with Grad-CAM heatmaps checked against clinical expectations, not just accuracy. Live demo. |
| AutoPrompt | Automated prompt optimization and benchmarking against a hand-written baseline — the project behind an accepted, co-authored paper. |
More context on all of these — including the decisions behind them, not just the pitch — is on the portfolio.
Languages — Python · TypeScript · JavaScript · Java Backend — FastAPI · Flask · PostgreSQL · MongoDB · Redis · Docker ML / AI — PyTorch · scikit-learn · Gemini & Groq APIs · RAG and agentic pipelines Frontend — React · Vite · Streamlit