TicketFairy

Landing page
Video gallery
Recording studio
Video and ticket preview
Project architecture

Inspiration

Writing tickets are like dentist appointments. Important, but never fun and almost always painful. They steal time, lose detail, and leave everyone wishing for a little magic. We wanted to make this experience feel like a visit from the Tooth Fairy.

With TicketFairy, anyone can turn a video into a feature. Whether you're a PM juggling 60 customer support requests, a developer refactoring the entire frontend for their startup, or even a user who found a bug in their favorite app, you can open TicketFairy, upload a screen recording, and watch as your work comes to life.

We help teams save hours of effort, cut miscommunication, and focus on execution.

What it does

TicketFairy enables teams to live capture or upload existing screen recordings, which are securely stored in the cloud with AWS S3 for easy access. The videos are then processed by a multimodal AI pipeline that combines multiple cutting-edge models and leverages accelerated computing for video understanding.

Using TwelveLabs' Pegasus model (twelvelabs.pegasus-1-2-v1:0), we leverage its low-latency video processing capabilities to generate contextually relevant text. After the indexing and open-ended analysis are complete, we apply custom prompts to extract comprehensive information for developer tickets and production-ready PRs.

Our goal is to provide flexibility and speed for modern product development workflows.

Using Cohere's Command R model (command-r-03-2024), we enhance and augment the various insights pulled from the video with optional user notes. This allows users to add additional context to their recordings like acceptance criteria and codebase structure to deliver quality responses.

TicketFairy integrates directly with tools like Jira and Linear and automatically creates tickets in the team's backlog. With context captured once, the rest of the process happens effortlessly using their respective APIs and AI-powered infrastructure.

Taking it to the next level, we introduced agentic workflows to our product. Using Claude Code and GitHub Actions, we let users integrate custom MCP servers and automate code implementation and PR creation. This means TicketFairy can interpret structured tickets as actionable development tasks and anyone, whether technical or non-technical, can generate a prototype for a requested feature or fix.

How we built it

Idea! ✨
Landing page + setup - React + Vite frontend, Flask backend
Screen recording feature with user webcam -> MediaRecorder API with webcam overlay
Connecting S3, uploading videos -> Cloud storage for video persistence
Compressing and processing video with TwelveLabs -> AI video analysis pipeline
Jira + Linear integrations -> Direct API connections for ticket creation
Agentic AI side quest that turned into main quest... -> Agent that automatically creates a Github action workflow that connects with Claude code agent that takes ticket and context from code base to put up a PR.
Prompt adjustments, formatting -> Custom prompts for structured ticket output
UI/UX polish -> Custom fonts, cursors, side-by-side video/ticket editor, loading states
Final refinements

Challenges we ran into

Video processing introduced latency, but our goal was to make ticket generation feel instantaneous
Video compression had to meet size constraints to work with API
Ensuring AI outputs were consistent, structured and useful
Running into token and usage limits
S3 permissions issues
Troubleshooting tool calls for GitHub integration
Permission blockers in workflow