AutoBot

Inspiration

We’ve all been in those chaotic incident calls where everyone’s talking, logs are flying, and no one remembers who said what. Writing postmortems afterward was painful. We wanted to build something that listens, understands, and documents everything automatically—so engineers can focus on solving problems, not writing reports.

What it does

AutoBot is our smart incident copilot. It listens to live meetings, turns conversations into structured insights, finds root causes, and publishes instant, customer-friendly updates—all while keeping sensitive data secure.

How we built it

We combined a FastAPI service for speech-to-text, a Node.js + Gemini LLM agent for reasoning and analysis, and Redis for orchestration. Integrations with Skyflow, Lightpanda, and Sanity handle privacy, automation, and real-time dashboards.

Challenges we ran into

Getting consistent, valid JSON from the LLM was tough. Handling data privacy across multiple services required careful planning. We also had to juggle many moving parts speech, AI, Redis, and APIs all working in sync.

Accomplishments that we're proud of

We built a full pipeline that goes from a meeting conversation to a published postmortem in minutes. Seeing it generate real, structured updates without human intervention was incredibly satisfying.

What we learned

We learned how powerful LLMs become when combined with well-designed schemas and event-driven architecture. And that real-time data privacy is not just possible it’s essential.

What's next for AutoBot

We want to take AutoBot further integrating with Slack, PagerDuty, and monitoring tools so it can automatically detect incidents and generate live reports before teams even join the call.

Built With

• lightpanda
• postman
• sanity
• skyflow