Building intelligent products from strategy to code.
UChicago MSADS · AI PM @ Meituan · Co-founder @ Eureka.ai
I work at the intersection of Generative AI, product strategy, and data.
My edge: I can define the spec and ship the demo.
Product focus areas:
- LLM product design, evaluation & performance tuning
- AI workflow automation
- 0→1 product launches in healthcare & consumer AI
Eureka.ai — AI-powered decision intelligence platform ↳ Source: github.com/khalil-he/Eureka — LangGraph + FastAPI + React, three-phase decision workflow with real-time canvas.
| Project | What I did | Impact |
|---|---|---|
| AI Quality Inspection (Meituan) | LLM-powered sales call QA — full spec + demo | −80% rule deployment cycle |
| AI Film Explorer (Meituan) | RAG + LLM ranking, owned AI modules | <3% hallucination rate |
| AI Tongue Diagnosis (Meituan) | PO for 0→1 launch, knowledge graph + vector search | 93% accuracy |
| Sony Competitive Intelligence | Multi-LLM automation workflow on Coze | 2hr → <5min daily |
| Repo | What it is | Stack |
|---|---|---|
| Eureka | AI strategic decision support — three-phase LangGraph workflow (Clarify → Solve → Synthesize) with a live decision-tree canvas | LangGraph · FastAPI · React · SSE |
| Deepfake-Bayesian-Detection | Deepfake detection that tells you when it doesn't know — calibrated uncertainty via MC Dropout & Variational Inference (team of 4) | PyTorch · Xception · Streamlit |
| RAG-based MSADS QA Agent | Local, citation-grounded QA agent for the UChicago MSADS site — DOM-aware knowledge graph + hybrid retrieval (vector + BM25 + structural) with a multi-step agent loop (team project) | Playwright · ChromaDB · BGE · Ollama (qwen3:8b) |
| KuaiRand Recommendation System | Collaborative-filtering benchmark on KuaiRand-Pure — Popularity / ItemCF / SVD / ALS, plus a two-stage ALS retrieval + GBT re-ranker pipeline (team project) | Python · ALS · GBT · Pandas |
LLM Product RAG Evaluation Frameworks Product Spec Next.js Python Data Analysis
