🔭 Currently building things that actually work: agentic ML pipelines, real-time CV systems, RAG chatbots, and the occasional "why does this even work?" project

👯 Down to collaborate on AI/ML projects, hackathons, or anything where data meets a real problem worth solving

🤝 Interested in learning from and working with people in advanced Machine Learning, MLOps, Deep Learning, and scalable AI systems

🔎 Deep in the MLOps rabbit hole right now: containerised pipelines, agentic deploy-gating, experiment tracking, and figuring out how to make agents that don't hallucinate every other sentence

💬 Hit me up about Python, ML pipelines, RAG systems, LangGraph, LangChain, data cleaning, or just bouncing around project ideas

⚡ Fun fact: I've built a 10-agent EDA pipeline, a RAG chatbot with zero API cost, a traffic system that actually predicts congestion, and a fully containerised agentic MLOps pipeline — still not satisfied 🚀

End-to-end agentic ML pipeline predicting next-day rainfall for New Delhi using live Open-Meteo API data. Orchestrated with Apache Airflow, tracked with MLflow (AUC 0.91, F1 0.66), gated by a LangGraph + Ollama (Llama 3.2) agent that reasons about deployment readiness, and served via FastAPI. Fully containerised — runs with a single docker compose up.

Airflow MLflow LangGraph Ollama Docker FastAPI Scikit-learn

Real-time traffic intelligence system using YOLOv8 for vehicle detection (92%+ accuracy), live signal timing optimisation (25% improvement), and a Flask dashboard with live control sliders. R²=0.89 on congestion prediction.

YOLOv8 Flask OpenCV Chart.js

10-agent EDA pipeline built with LangGraph that automates exploratory data analysis — 80% time reduction, 15+ visualisations, supports 100k+ row datasets.

LangGraph LangChain Pandas Matplotlib

Zero API-cost RAG chatbot with PDF and YouTube ingestion, FAISS vector search, MiniLM embeddings, and configurable chunk/overlap/top-K sliders via Streamlit.

LangChain FAISS Ollama Streamlit

Hybrid movie recommender combining SVD matrix factorization and TF-IDF genre similarity on MovieLens Latest Small (9,700 movies). RMSE 0.871, Precision@10 55.8%, Recall@10 66.2%. Interactive Streamlit dashboard with latent factor visualizer, α sensitivity heatmap, and user similarity explorer.

SVD TF-IDF Scikit-surprise Streamlit Plotly