thai_llm_kmitl

เจ้าของผลงานเล่าไว้แบบนี้

https://thai-llm-kmitl.vercel.app # thai_llm_kmitl RAG chatbot for the **KMITL Faculty of Information Technology** Open House: answers high-school students' questions about the four B.Sc. programs (AIT, DSBA, BIT, IT) from the official curriculum documents. Competition rule: **only ThaiLLM models** may read, analyse or answer anything at runtime (chat completions only, no embeddings). ## Quickstart ```bash uv sync # Python >= 3.12; installs runtime + dev deps cp .env.example .env # fill in THAILLM_API_KEY uv run pytest -q # pure-function + mocked tests, no network uv run ruff check # lint ``` Run the API (stub answerer — real event shapes, fake answer, no retrieval): ```bash ANSWERER=stub uv run uvicorn api.main:app --reload # http://localhost:8000 ANSWERER=rag RETRIEVER=fixture uv run uvicorn api.main:app --reload # real answer layer over 120 real curriculum passages (tests/fixtures/chunks.jsonl) curl -N localhost:8000/chat -H 'content-type: application/json' \ -d '{"message": "หลักสูตร AIT เรียนกี่หน่วยกิต"}' curl localhost:8000/health ``` ## Run the full stack locally Frontend (`web/`, Next.js) → `POST /api/chat` (Next route) → FastAPI `POST /chat` (gate → RAG, SSE). One command (starts both, Ctrl-C stops both, logs in `.cache/dev/`): ```bash scripts/dev.sh # ANSWERER=rag RETRIEVER=fixture on :8000, Next.js on :3000 ``` Or two terminals: ```bash # terminal A — backe

โดย LoveMig6334 · ดูโพสต์ต้นทาง