Loading…
A fully local RAG and multi-agent system over scraped Samsung phone specifications. Pick any of the 188 labelled evaluation questions to see how it was routed, which passages it retrieved, what it answered, and whether the answer was correct.
These are recorded outputs of the real pipeline (after the routing fixes), not a live model: hosting a live Python app here needs a paid plan. To run it live, clone the repository and use Docker or python scripts/run_api.py. Template answers are deterministic; LLM answers (Qwen2.5-1.5B, test split only) were generated once on a laptop GPU.
Loading…