MicroManager

About MicroManager

MicroManager was inspired by a problem we kept running into during team projects: meetings create a lot of discussion, but the actual execution often gets lost afterwards. Someone has to manually write tasks, remember who is responsible, check deadlines, follow up with people, and notice blockers before they become serious. We wanted to build an AI assistant that does that automatically.

MicroManager turns a meeting transcript into a live project dashboard. A user can paste meeting notes or a transcript, and the AI extracts tasks, owners, deadlines, dependencies, follow-ups, human decisions, and project risk. Instead of leaving action items buried in conversation, MicroManager converts them into something the team can actually track.

How We Built It

We built MicroManager using a simple web stack so we could move quickly during the hackathon:

Frontend: HTML, CSS, and JavaScript
Backend: Python with FastAPI
Database: SQLite
AI: Gemini API for transcript understanding
Authentication: Firebase Google login

The backend processes the transcript, sends it to the AI model, receives structured project data, and stores it in SQLite. The frontend then displays this across pages for tasks, personal assigned work, people, follow-ups, task meetings, project risk, decisions, and agent activity.

What We Learned

We learned a lot about turning AI output into a real product experience. It is not enough for the AI to simply summarize a meeting. The difficult part is making the output structured, reliable, editable, and useful across the whole app.

We also learned how important small product details are, like matching tasks to the logged-in user, showing why a project is risky, and making the dashboard feel like an actual assistant rather than just a static task list.

Challenges We Faced

One of the biggest challenges was getting consistent structured responses from the AI. Sometimes the model returned extra text, invalid JSON, or slightly different formats, so we had to make the backend more robust.

We also faced issues with API setup, local backend connection problems, Firebase authentication, and keeping the frontend synced with the database. Another challenge was deciding what should be fully automated and what should still stay human-controlled, especially for decisions and follow-ups.