Student learning AI & Machine Learning · building ML and LLM projects end to end · Contributor at Cylstat
A few things I've built to learn by doing - from classic ML to LLM apps, several with live demos you can try.
Capstone: an internal assistant combining RAG over documents and a tool-using agent that queries a SQL database. FastAPI + Streamlit + Docker.
Abstractive summariser (distilBART) that handles long documents via chunking. Python package, CLI and a live web demo.
Transfer-learning ResNet-18 recognising five flower species. Model on the Hub and a live demo where you upload a photo.
Retrieval-augmented generation over a document corpus: chunking, embeddings, cosine search and grounded answers with citations.
A tool-using agent that answers natural-language questions over a SQL database, with a read-only query guard.
End-to-end ML pipeline (EDA → preprocessing → model comparison → evaluation) on the Telco dataset. Also a Kaggle notebook.
A small harness to compare an LLM against baseline classifiers on a labelled task, with consistent metrics.
CLI toolkit for regular expressions: explain (AST), test, profile (ReDoS detection) and audit/mask PII, with an interactive shell.