Projects

Here are a few projects that represent the kind of work I enjoy most: practical ML systems, LLM tooling, and end-to-end pipelines with clear evaluation. Each one links to the repo so you can see the code, documentation, and experiments.

Project 1: GPT-2 IMDB fine-tuning + evaluation toolkit

Fine-tuned GPT-2 on the IMDB dataset and built a small evaluation toolkit to measure quality consistently across runs. Includes preprocessing, training scripts, experiment tracking, and simple metrics reporting so results are easy to reproduce. GitHub

Project 2: Conversational RAG system (LangChain/LangGraph + FAISS + SentenceTransformers/Ollama)

Built a conversational RAG chatbot with memory and document retrieval using LangChain/LangGraph. Uses FAISS for fast similarity search and embedding models (SentenceTransformers/Ollama) to keep responses grounded in your knowledge base. GitHub

Project 3: Fraud / classification pipeline work (metrics, PR-AUC, cost-based thresholding)

Created a fraud classification pipeline focused on real-world evaluation: PR-AUC, precision/recall tradeoffs, and cost-based thresholding. Includes feature engineering, model training, and a clear metrics workflow for choosing operating points. GitHub

Project 4: Deepfake detection using ELA + CNN

Implemented a deepfake detection approach combining Error Level Analysis (ELA) with a CNN classifier. Covers image preprocessing, model training, and evaluation — with emphasis on building a repeatable pipeline rather than a one-off notebook. GitHub

Project 5: Healthcare LLM evaluation framework (NIST/Stanford-aligned metrics)

Designed an evaluation framework for healthcare-style LLM tasks with an emphasis on scoring, consistency, and usefulness. Includes metric-driven analysis (inspired by common research/industry evaluation ideas) and a structure for expanding tasks over time. GitHub

Project 6: Portfolio website

Built this portfolio site to showcase projects and make it easy to contact me. It’s responsive, lightweight, and deployed with a clean workflow so updates are quick and reliable. GitHub