Projects

JLG Industries (Oshkosh)

Production internal assistant using vector search + clustering over embedded internal docs. Role-restricted spaces ensure precise, compliant answers for each department.

• Cut knowledge lookup time across departments by hours/week per user
• Higher answer precision via department-aligned embedding spaces
• Pipeline prepared for scale (ingestion → embedding → retrieval → evals)

JLG Industries (Oshkosh)

End-to-end ML + LLM flow for routing emails, summarizing threads, and proposing context-aware replies that users can one-click approve.

• Reduced manual triage time; consistent routing & responses
• Unified pipeline from raw inbox → model → user interface

Next.js + Tailwind portfolio with a custom chat widget backed by a FastAPI service on Railway. Uses Llama-3.1-8B-Instruct for responses, all-MiniLM-L6-v2 for embeddings, and FAISS for fast vector search. The assistant performs semantic retrieval over my resume/notes to answer recruiter questions.

• Interactive, grounded answers about my experience (semantic retrieval > keyword search).
• Clear separation of concerns: frontend on Vercel, backend on Railway with CORS/secrets configured.
• Low-latency retrieval and stable responses using FAISS + carefully constrained prompts.