File size: 2,234 Bytes
58b1106 630f6f1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 | ---
title: Phuree Healthcare Assistant
emoji: 🏥
colorFrom: blue
colorTo: green
sdk: docker
app_port: 7860
pinned: false
---
# Qwen Healthcare Assistant
This is a healthcare assistant powered by a custom Qwen model.
## Current Runtime Flow
The Docker app runs `main_v2:app` with FastAPI and exposes:
- `GET /` for a health check.
- `GET /generate?query=...&useRAG=false&thread_id=...` for SSE chat streaming.
- `GET /clear?thread_id=...` to clear one conversation thread.
### Request Path
```text
Client
-> GET /generate?query=...&useRAG=...&thread_id=...
-> FastAPI generate_endpoint()
-> Load LangGraph memory by thread_id
-> If new thread: add initial health-assistant system message
-> If useRAG=true: add internal force-retrieval marker
-> Add user query
-> Run LangGraph
-> Stream SSE response back to client
```
### Routing Path
```text
User query
-> Pure small talk?
-> Yes: direct short reply
-> No:
-> Health capability/meta question?
-> Yes: direct scope reply
-> No:
-> Clear animal or pet-health question?
-> Yes: direct outside-scope reply
-> No:
-> Clear human-health question?
-> Yes: retrieve_health_info tool call
-> No:
-> LLM router classifies query
-> RETRIEVE: retrieve_health_info tool call
-> ANIMAL: direct outside-scope reply
-> DIRECT: direct response with conversation history
```
### RAG Path
```text
retrieve_health_info
-> Chroma vector search
-> BAAI/bge-m3 embeddings
-> Candidate documents
-> BAAI/bge-reranker-v2-m3 cross-encoder rerank
-> Top reranked documents
-> RAG prompt with relevant context
-> Qwen/Ollama answer
-> Clean <think>...</think>
-> Stream:
**Source:**{...}
final answer
```
### Direct Path
```text
DIRECT route
-> Build prompt from:
DIRECT_RESPONSE_SYSTEM_MESSAGE
+ prior human messages
+ prior AI messages without tool calls
-> Qwen/Ollama answer
-> Clean <think>...</think>
-> Stream final answer
```
### Error Path
```text
Ollama / router / RAG / streaming failure
-> Catch exception
-> Log error on server
-> Stream fallback message to client
-> Avoid raw 500 traceback in chat UI
```
|