PlantAI-rag/docs/plans/PROGRESS.md

Gene Family MAS Project Progress

Backup Location

• /home/xia-lab/gene-family-mas-backup-20260221-053220

Papers Location

• /home/xia-lab/wutx/00test/bbav2/paper/gfpaper/buchong/ (33+ PDFs)

Completed Modules (2026-02-21)

1. LLM Multi-Provider System (src/llm/)

• base.py: BaseLLMProvider, LLMResponse, LLMMessage
• manager.py: LLMManager with auto-fallback (GLM-4 → Kimi → MiniMax → DeepSeek)
• providers/glm.py: GLM-4 via zhipuai
• providers/kimi.py: Kimi via OpenAI-compatible API
• providers/minimax.py: MiniMax via OpenAI-compatible API
• providers/deepseek.py: DeepSeek via OpenAI-compatible API

2. Memory System (src/memory/)

• base.py: BaseMemory, MemoryEntry, SearchResult
• session.py: SessionMemory (in-memory with TTL)
• case.py: CaseMemory (SQLite persistence)
• knowledge.py: KnowledgeMemory (vector embeddings)
• manager.py: MemoryManager orchestrating all layers

3. Core Orchestrator (src/core/)

• state.py: GraphState (TypedDict for LangGraph)
• orchestrator.py: LangGraph flow with intent routing (FAQ/Analysis modes)

4. RAG Knowledge System (src/rag/) ✅

• base.py: Document, Chunk, SearchResult dataclasses
• pdf_extractor.py: PDF text extraction using PyMuPDF
• chunker.py: 50-30 chunking strategy
• embeddings.py: GLMEmbedding with fake embedding fallback
• vector_store.py: In-memory vector store with cosine similarity
• retriever.py: RAGRetriever orchestrating extraction → chunking → embedding → search

5. Tool Execution System (src/tools/) ✅

• registry.py: ToolRegistry with HMMER, BLAST, MAFFT, IQ-TREE, MCScanX, MEME
• adapter.py: ToolAdapter for command execution
• executor.py: ToolExecutor with Docker support
• workflow.py: GeneFamilyWorkflow for chaining tools

6. FastAPI Backend (api/)

• main.py: FastAPI app with CORS, dotenv loading
• routes.py: Full REST API endpoints
  • /session, /query, /case, /feedback
  • /providers - List available LLM providers
  • /tools - List available analysis tools
  • /tools/{tool_id}/execute - Execute a specific tool
  • /workflow/gene-family - Run gene family workflow
  • /upload - Single file upload
  • /upload/batch - Batch file upload
  • /ws/{session_id} - WebSocket for real-time progress
• schemas.py: Pydantic request/response models
• websocket.py: WebSocket connection manager

7. Frontend (frontend/)

• Next.js 14 + TypeScript + Tailwind CSS
• Chat interface with message history
• File upload support
• WebSocket for real-time progress

API Endpoints Summary

Endpoint	Method	Description
/health	GET	Health check
/api/v1/providers	GET	List LLM providers
/api/v1/session	POST	Create session
/api/v1/query	POST	Submit query (FAQ/Analysis)
/api/v1/case	POST/GET	Create/get case
/api/v1/feedback	POST	Submit feedback
/api/v1/tools	GET	List available tools
/api/v1/tools/{tool_id}/execute	POST	Execute tool
/api/v1/workflow/gene-family	POST	Run analysis workflow
/api/v1/upload	POST	Upload single file
/api/v1/upload/batch	POST	Upload multiple files
/api/v1/ws/{session_id}	WS	WebSocket connection

Configuration

Environment Variables (.env)

ZHIPUAI_API_KEY=configured
MOONSHOT_API_KEY=configured
MINIMAX_API_KEY=configured
DEEPSEEK_API_KEY=configured

LLM Config (config/llm.yaml)

• Default: GLM-4-flash
• Models: glm-4-flash, glm-4, glm-4-plus
• Fallback: Kimi → MiniMax → DeepSeek

Running Commands

Backend

cd /home/xia-lab/gene-family-mas
poetry shell
uvicorn api.main:app --host 0.0.0.0 --port 8000

Frontend

cd /home/xia-lab/gene-family-mas/frontend
npm run dev

Test Queries

# FAQ query
curl -X POST http://localhost:8000/api/v1/query \
  -H "Content-Type: application/json" \
  -d '{"session_id": "test", "query": "什么是基因家族？"}'

# Analysis query
curl -X POST http://localhost:8000/api/v1/query \
  -H "Content-Type: application/json" \
  -d '{"session_id": "test", "query": "帮我鉴定苦荞的SPL基因家族"}'

# File upload
curl -X POST http://localhost:8000/api/v1/upload \
  -F "file=@sequences.fasta"