Minute – Retrieval-Based AI Meeting Co-Host for BFSI
VNPT AI Hackathon 2025 | Desktop/Web meeting workflow with citations and audit trails
Minute standardizes the meeting lifecycle for BFSI/LPBank enterprises: pre-meeting context gathering, real-time in-meeting assistance, and post-meeting minutes + action items generation, with citations, audit trails, and access control. The system uses stage-aware routing, a real-time WebSocket pipeline, permission-aware retrieval with pgvector, and human confirmation for tool actions.
Timeline
2025
Type
Competition
Status
completed
My work
- •Worked alongside one other code contributor on the Minute implementation
- •Designed the pgvector data model, document-ingestion flow, and metadata/ACL filtering for permission-aware retrieval
- •Contributed to LangGraph routing, FastAPI APIs, and the real-time WebSocket meeting pipeline
Outcome / Impact
- •Built end-to-end AI meeting workflow: Pre-meeting (agenda + pre-read) → In-meeting (live transcript, recap, ADR extraction) → Post-meeting (executive summary, MoM, task sync)
- •Contributed to stage-aware LangGraph routing for the Pre/In/Post meeting workflow
- •Contributed to the real-time WebSocket pipeline with SmartVoice STT integration
- •Data modeling with pgvector embeddings, document ingestion (OCR/SmartReader), and optimized RAG queries
Tech / Skills
Project Media
Demo video and visual walkthrough for this project.
Demo Video
Open source videoCertificates (1)
VNPT AI Hackathon 2025 - 3rd Prize, Track 1
Case Study
1) Context / Problem
Enterprise meetings in BFSI consume significant time but often fail to produce clear decisions, accountable action items, and traceable outcomes. Manual minute-taking is slow, error-prone, and lacks audit trails. LPBank needed a solution that integrates with existing workflows while meeting strict compliance requirements.
2) Your Role
As AI Engineer, I was a core contributor to the full-stack implementation. I designed the data model and pgvector schema, built the document-ingestion pipeline (OCR/SmartReader), optimized RAG queries with metadata filtering, contributed to LangGraph orchestration, and collaborated on backend API and real-time WebSocket features.
3) Approach
Built SAAR (Self-aware Adaptive Agentic RAG) architecture with: (1) Stage-aware LangGraph router for Pre/In/Post meeting phases, (2) Real-time WebSocket pipeline for audio → STT → session bus → live transcript, (3) Permission-aware RAG with pgvector + ACL filters, (4) Tool-calling with human-in-the-loop for task creation and scheduling.
4) Result / Impact
Delivered a Minute demo with end-to-end Pre/In/Post workflow, real-time transcription with ADR extraction, grounded RAG Q&A with citations, and audit-ready structured outputs. The team won 3rd Prize in VNPT AI Hackathon 2025, Track 1.
5) Learnings
Learned to balance real-time latency requirements with LLM quality, implement effective graded retrieval strategies, and design for enterprise compliance. Would explore streaming LLM responses and more aggressive caching for production.
6) Links
See links above.