web
UI Layer
Streamlit chat UI captures the prompt, builds GraphState, invokes LangGraph, and renders responses plus upload controls.
Agentic AI Capstone
Enterprise Assistant
End-to-end enterprise assistant built with Streamlit + LangGraph that combines retrieval Q&A, operational tool automation, and RFP proposal generation in a single routed architecture.
hub
Architecture At A Glance
account_tree
Orchestration Layer
Supervisor classifies intent and routes to RAG_FLOW, TOOL_FLOW, RFP_FLOW, or UPLOAD_FLOW using conditional graph edges.
smart_toy
Execution Layer
Specialized agents execute retrieval Q&A, SQLite tool actions, enterprise RFP drafting, and upload-first workflows.
storage
Data + Ops Layer
Retrieval runs on FAISS or Qdrant, tool data comes from SQLite, and standardized logger traces each request_id end to end.
Request Lifecycle
1) Receive Question
Streamlit captures user text with st.chat_input(), generates request_id, and initializes GraphState with optional uploaded_file.
2) Classify Intent
Supervisor node in nodes.py calls router.py intent logic with keyword-first routing and optional LLM fallback.
3) Route by Graph
graph.py sends flow to rag/tool/rfp/upload nodes through conditional edges and then exits at END.
4) Execute Agent
RAG uses similarity search context, TOOL queries open incidents/requests, RFP drafts proposals with optional PDF context.
5) Render + Log
Response, intent, and upload hints are returned to UI; chat history updates and logs include flow decisions and response preview.
What Is Built
- LangGraph multi-route architecture with supervisor + deterministic routing paths.
- RAG pipeline with local FAISS indexing and cloud Qdrant retrieval abstraction.
- Tooling agent integrated with SQLite incidents and service request tables.
- RFP proposal generator using uploaded PDF parsing with bounded context extraction.
- Provider-aware model layer supporting Ollama, Gemini, and Groq via config.
- Observability-first logging chain from UI events to node and tool execution.
How To Use
- Ask operations queries: `show open incidents`, `show open requests`.
- Ask enterprise drafting prompts: `Draft proposal for cloud migration`.
- Use upload flow: `I want to upload an RFP` then attach PDF and request summary/draft.
- Run app locally with `poetry run streamlit run app/ui/streamlit_app.py`.
- Initialize data with `poetry run python -m app.db.init_db` and `poetry run python -m app.db.seed_db`.
- Build retrieval index with `poetry run python -m app.rag.build_index` (local) or ingest Qdrant for cloud mode.