Final Project
Showcase

A 4-agent LangGraph pipeline that turns a Pinterest-style vision board into ranked, live product results from real retailers.

Builds therapeutic playlists for dementia patients by grounding retrieval in the reminiscence bump, region, and life events.

Teaches small language models to reliably call APIs through SFT with function masking, GRPO, and easy-to-hard curriculum learning.

Agent-based simulation of Berkeley Hills homeowners deciding whether to harden their properties under shifting fire-risk policy.

Personalized AI coaching for AuDHD users — task decomposition and reminders shaped by prior plans and neurodivergent-focused prompting.

A real-time conversational recommender that fuses live restaurant data with community sentiment to rank suggestions on the fly.

Generates film marketing taglines and poster art grounded in TMDB retrieval and sharpened by an LLM-as-Judge refinement loop.

Turns raw interview transcripts into traceable user stories and pain points, comparing ChromaDB vs. FAISS retrieval on the same corpus.

A copyright-law research assistant that pits a FAISS-backed RAG pipeline against in-context case briefs — with human evaluation on both.

A multi-agent researcher that scrapes live listings, cross-references market signals, and synthesizes investor-ready rental intelligence.

Nutrition tracking plus constraint-aware meal planning — respects macros, allergies, and pantry state when suggesting what to eat next.

A 5-modality RAG system — text, tables, images, audio, and video — unified so users can query any document type with one question.