Proposal AI Generator
Multi-Agent System
Slack-native proposal automation · built in one week
LangGraphGroq (LLaMA 3.3 70B)QdrantSlack SDKFastAPISQLitepython-docxStreamlit
- Built a three-agent pipeline that converts client discovery call transcripts into fully formatted DOCX business proposals, delivered directly in Slack within seconds of upload.
- Orchestrated stateful agent handoffs with LangGraph: one agent extracts structured client data, one retrieves relevant context via RAG, and one synthesizes the final proposal.
- Implemented semantic retrieval over Qdrant with industry metadata filtering, so retail clients surface retail-relevant past proposals.
- Designed composite session keys (user_id:channel_id) for concurrent multi-client proposal management with cross-session memory.
- Added targeted section-level editing — follow-up requests update only the relevant section and redeliver the revised DOCX automatically.
