As a Business Analyst and Product Owner, I spend days after every meeting turning discussions into requirements, backlogs, test cases, traceability matrices, and diagrams. It's slow, repetitive, and easy to miss things. I wanted to close the gap between "discussion" and "delivery." What it does ? Discussion → Delivery is a multi-agent AI hub that turns any meeting transcript into a complete delivery package in one click. Seven specialized OpenAI-powered agents work together:
- BA Agent – Business Requirements Document with scope, risks, priorities, confidence scores and acceptance criteria
- PO Agent– product vision, epics and prioritized user stories
- QA Agent – positive, negative, edge and UAT test cases
- Governance Agent – full traceability matrix linking requirements → stories → tests
- Architecture Agent** – process, swimlane, architecture and data-flow diagrams, exportable to Draw.io/Visio
- Focus Agent** – plans the user's day and summarizes what's done and pending It also supports multi-session input, merging several meetings and removing duplicates. How we built it ? Built with a Python FastAPI backend and a React frontend, using the OpenAI API for all seven agents. Each agent has structured prompts that return JSON, which the UI renders into professional artifacts. Diagrams are generated dynamically and exported to Draw.io/Visio. Challenges we ran into Keeping AI output structured and reliable, handling long or multi-session transcripts within token limits, and making the tool honest — every requirement carries a confidence score and anything uncertain is flagged "Needs Validation" so a human stays in the loop. What we learned ? That the goal isn't 100% automation — it's accurate drafts plus honest flags. Good prompts and human review matter more than replacing the human. What's next ? Outlook/calendar integration for the Focus Agent, central hosting, role-based access, session analytics to measure ROI, and prompt refinement for even higher accuracy. Impact Turns days of BA work into minutes — an estimated ~20 hours saved per sprint per BA (~480 hours/year), serving the whole delivery team from a single meeting.
Built With
- agentic-ai
- ai-agents
- business
- css
- drawio
- fastapi
- gpt
- gpt-4o-mini
- html
- javascript
- mermaid
- multi-agent
- openai
- process
- python
- react
Log in or sign up for Devpost to join the conversation.