MACHINELEARNINGENGINEER

Two-Tower Recommender

End-to-end retrieval and ranking. A two-tower model learns dense user and item embeddings, FAISS handles candidate generation, and a PyTorch MLP reranks the shortlist. Beats matrix factorization and GBDT baselines on Recall@50, NDCG@10, and MRR.

Agentic LLM Red-Teaming

An eight-node LangGraph agent that takes a deployment spec, generates adversarial attacks, judges every response against a seven-verdict rubric, and clusters failures into families with mitigations. Against a gemini-2.5-flash bank assistant on a 30-prompt HarmBench + XSTest suite, it caught every harmful request but surfaced 12 safe-question over-refusals traced to one fixable prompt rule.

SportsIQ Reasoning Agent

A ReAct agent over three domain-partitioned corpora (rules, tactics, game situations) with hybrid dense and sparse retrieval plus cross-encoder reranking, on the hardest tier of SportQA. A confidence-gated, hop-routed ensemble reaches 89.5% token-level F1 and 67.1% exact match, beating a no-RAG GPT-5.4 baseline, with 99.8% retrieval precision@k.

Team project with Cady He and Siyani Vengatagiri.

LLaMA-Style Transformer From Scratch

Scaled dot-product and grouped-query attention, RoPE, SwiGLU, LayerNorm, top-p sampling, and AdamW, all implemented from scratch in PyTorch. Training and eval pipelines cover text generation, zero-shot sentiment on SST and CFIMDB, and arithmetic reasoning.