The Problem
Kioku was built for companies and universities to automate meeting minutes and organize meeting activities, including invitations, attendance, and distributing minutes.
My Contribution
As project lead and backend developer, I implemented meeting creation, email invitations, login-free magic-link check-in, manual attendance, recording uploads, and recording document management. I also developed automatic email distribution of processed minutes and search across meetings and minutes.
Process and Decisions
The AI workflow takes uploaded recordings through transcription, speaker diarization, extraction of decisions and action items, and meeting minutes generation. We initially ran all AI models locally. When limited computing resources made processing too slow, I added options to use cloud models from Gemini and OpenAI.
Challenges
This was my first project using Docker, which introduced a learning curve in configuring the application environment. I also learned to handle audio processing asynchronously with Celery and Redis so the application could continue serving requests while processing recordings.
Results
The system is complete but has not yet been deployed. It supports recording uploads of up to two hours, meeting and attendance management, and automated minutes distribution. Google Calendar integration and SSO login have also been implemented.
Reflection
Kioku taught me how audio processing fits into a complete application workflow. I gained practical experience with Docker, Redis, and background tasks, while learning to adapt technical choices to available computing resources.