2026 · Agentic AI
UdaPlay Research Agent
A stateful agent that checks memory and a vector store, judges retrieval quality, then falls back to web search.
- Question
- Memory and retrieval
- Grounded response
Complete portfolio project (Udacity capstone, refactored)GitHub ↗
Problem
Answer video-game industry questions reliably, knowing when local knowledge is not enough.
My role
Author of the repository.
Approach
- Check long-term JSON memory, then retrieve from a ChromaDB store of 15 curated game records.
- An LLM-as-judge returns a structured evaluation (confidence, usefulness, needs web search).
- Route to a local answer, memory reuse, or Tavily web search; save web answers to memory; return cited, confidence-graded structured output.
Results
- Tests run with mocks and need no API keys. No quantitative accuracy results are published.
Stack
- Python
- OpenAI
- ChromaDB
- Tavily
- Streamlit