Physics graduate, self-taught into AI/ML engineering through five shipped systems — not a stack of course certificates. Each one is deployed, evaluated, and documented the way a lab notebook records a result: what was built, how it was tested, what it proved.
I didn't come up through a CS degree. I studied Physics at the University of Mumbai — where the job isn't to have an opinion, it's to design an experiment, run it, and let the result speak. That habit is the throughline in everything I build now: not "does the demo look good," but "does it hold up when you measure it."
After a year spent exploring direction, I enrolled in a structured Data Science & AI program at the Boston Institute of Analytics in December 2025 — and in parallel, taught myself the rest by shipping. Five end-to-end systems, each one deployed and evaluated, not left in a notebook.
I have no internship on my resume yet. What I have instead is a body of work I can walk anyone through line by line — architecture, failure cases, and the numbers that prove it works.
An entry-level role at an analytics-focused or AI-native team — the kind of place where a working system and a clear evaluation matter more than which degree is printed on my resume.
I'm equally interested in roles titled "Data Scientist" or "GenAI Engineer" as I am in "AI/ML Engineer" — the work overlaps more than the titles suggest, and my projects span both the modeling side (fraud detection, feature engineering) and the agentic/LLM side (RAG, tool-calling, multi-agent evaluation).
This site exists to do one thing: let a hiring manager verify, in under three minutes, that I can take an AI system from architecture to a deployed, evaluated result. Not a portfolio of promises — a log of experiments that passed.
An agentic assistant with LangChain tool-calling for search, recommendations and checkout, scored by an automated LLM-as-a-Judge pipeline for hallucination and format compliance, then containerized and deployed on AWS EC2/ECR.
An end-to-end retrieval pipeline over fragmented support content using ChromaDB and sentence-transformer embeddings, with similarity-threshold gating and inline source citations — measured at 100% Recall@3 on a curated benchmark query set.
A tuned Gradient Boosting pipeline over 11K+ transactions, built to minimize false negatives in financial identification, monitored through a Streamlit app and a 6-visual Power BI dashboard for real-time auditability.
A production-accessible full-stack app with JWT-authenticated, session-scoped endpoints, architected across three independently hosted services — debugged through real CORS conflicts and driver incompatibilities to get it live.