Ava - Agile Virtual Assistant

Inspiration

We noticed a disconnect between current technologies' capabilities and the daily realities of Agile development. We saw that the "Agile tax"—the administrative work of ceremonies, tracking, and long meetings—was a drag on our team's velocity and focus. We were inspired to build a tool that handles this 'work about work' so developers can get back to building.

What it does

AVA is an AI Scrum Master designed to join your team and improve your Agile methodologies. She records meetings, generates instant meeting minutes, and provides personalized, constructive feedback to team members. AVA also manages processes by tracking action items, keeping ceremonies on topic, and automatically creating custom Confluence pages from your meeting data.

How we built it

AVA is built primarily in Python, using discord.py to power her interactions as a Discord bot. We structured the internal logic around PydanticAI, defining strict Pydantic models to orchestrate prompts, services, and feedback flows. To handle meeting transcripts, we integrated the Google Drive API. Since webhook push notifications were unreliable, we implemented a polling system that continuously checks for file changes, automatically downloads new transcripts, parses them, and generates summaries and feedback. The backend services are modular, type-safe, and easy to extend thanks to Pydantic models and FastAPI-style design patterns.

Challenges we ran into

One of the major challenges was trying to integrate AVA with Microsoft Teams. We spent a significant amount of time setting up Azure Bot resources, configuring endpoints, and navigating permissions—but hit multiple blockers that made a working integration infeasible within our timeline. We also attempted to use Google Drive webhooks to monitor transcript uploads, but due to limitations with verification tokens and the webhook setup, this approach wasn’t reliable. In the end, we pivoted to a polling architecture, which allowed us to move forward without being blocked.

Accomplishments that we're proud of

Even with blockers, we successfully pivoted from Teams to Discord and built a fully functioning Agile assistant. We’re proud of the LLM-generated feedback quality, which turned out significantly stronger than expected thanks to careful prompt engineering, structured role/context modeling, and clean Pydantic-based pipelines. We were able to generate clean, informative confluence pages with enterprise level standards.

Whats next for Ava

Our next steps include completing a true Microsoft Teams integration so AVA can join organizations in their native workflow. We also want to build a custom transcript ingestion pipeline, giving us full control over audio processing, summarization, and action-item extraction—removing external dependencies and making AVA fully end-to-end.