How anilkumarb20062006/project works
This repository tree combines several projects, but its README identifies the Zepto Support Assistant as the product: a FastAPI question-answering service grounded in policy documents. The main workflow accepts a question, classifies it, retrieves policy context for policy questions (or gives a fallback for general questions), and returns a structured answer. The README also describes document ingestion and an optional real-LLM extension. The tree places policy files at its root while the sampled vector-store code expects a docs/ directory, so the ingestion-to-store link is uncertain; the optional LLM functions are explicitly unimplemented.
Main entry points: main.py Ask API · graph.py Intent classifier · vectorstore.py Vector store · prompt_template.py Prompt builder
Components 13
API boundary
- Ask API
main.py - Request/response schema
schema.py
Answer workflow
- LangGraph workflow
graph.py - Intent classifier
graph.py - Policy answer
graph.py - General fallback
graph.py - Prompt builder
prompt_template.py
Policy knowledge
- Policy corpus
- Vector store
vectorstore.py - ChromaDB collection
vectorstore.py - Local embedder
vectorstore.py
Other components
- Questioner
- Real LLM
Connections 15
- Questioner → Ask API: submits question
- Ask API → Request/response schema: validates input
- Ask API → LangGraph workflow: invokes workflow
- LangGraph workflow → Intent classifier: starts at
- Intent classifier → Policy answer: routes policy
- Intent classifier → General fallback: routes general
- Policy answer → Vector store: queries top-k
- Vector store → Local embedder: embeds query
- Vector store → ChromaDB collection: searches collection
- Policy answer → Prompt builder: builds optional prompt
- General fallback → Prompt builder: builds optional prompt
- Intent classifier → Real LLM: optional classification
- Prompt builder → Real LLM: optional generation
- LangGraph workflow → Ask API: returns answer state
- Ask API → Request/response schema: constructs response