Pit-Stop Inspect Agent

For Williams Racing: Stop guesswork. Rovo's Pit-Stop Agent delivers AI-driven certainty on F1 defects, instantly turning damage into actionable Jira tasks. Go faster, win more.

Comment

Inspiration

Inspired by this year’s theme, "Unleash your elite racing crew," we looked at where racing teams actually lose time: the pit stop and the post-race inspection. In a high-pressure environment, human mechanics are incredibly skilled but can be overwhelmed by data or fatigue.

We asked ourselves: What if the pit crew had an AI-powered chief mechanic that never blinks? We wanted to build a tool that doesn't just "chat" but actively works alongside the crew. Seeing what they see (via images) and knowing what they know (via Confluence specs) to automate the tedious administrative work of logging repairs in Jira.

What it does

Pit-Stop Inspect Agent is a specialized Rovo Agent that streamlines vehicle damage assessment and repair workflow management.

  • Visual Analysis: The user provides images of the car (pre- and post-race).
  • Context-Awareness: The Agent fetches specific vehicle technical data (e.g., required tire tread, wing dimensions) directly from a Confluence page.
  • Smart Detection: It compares the visual evidence against the Confluence specs to detect damages or deviations.
  • Actionable Output: Instead of just listing errors, it acts. It creates a parent Jira issue for the inspection and automatically generates individual Sub-tasks for each detected damage, ready for the repair crew to tackle.

How we built it

We utilized the Atlassian Forge platform to build a Rovo Agent with specialized capabilities:

  1. Rovo Agent: We configured the Agent with a specific system prompt designed for "Elite Maintenance" personas, focusing on precision and brevity.
  2. Confluence API: We implemented a function that allows the Agent to retrieve content from Confluence pages via Page ID, using this as the "Ground Truth" for the inspection.
  3. Multimodal Analysis: We leveraged Rovo’s underlying capability to process image inputs alongside text instructions.
  4. Jira REST API: We built a backend module that takes the structured output from the AI analysis and orchestrates the creation of Jira Issues and Sub-tasks, ensuring the correct hierarchy.

Challenges we ran into

  • Context Grounding: It was challenging to get the Agent to strictly compare the image against the Confluence data, rather than just hallucinating general car problems. We had to refine the prompts to ensure it treats the Confluence page as the absolute standard.
  • Structured Output: Getting a conversational AI to output clean data that could be programmatically turned into Jira tickets (titles, descriptions, priorities) required several iterations of prompt engineering and function calling definitions.
  • Image Consistency: Handling different angles and lighting in user-uploaded photos proved tricky for consistent damage detection.

Accomplishments that we're proud of

  • Runs on Atlassian: We build the app only using native Forge features without leaving the Atlassian ecosystem and therefore meeting the highest privacy and security standards.
  • Seamless Workflow: We are proud of how smooth the flow feels. You stay in the chat, upload a picture, and seconds later, real work items appear in the Jira backlog.
  • The "One-Click" Setup: Successfully turning a natural language conversation about a car crash into a structured list of Jira Sub-tasks felt like magic the first time it worked.
  • True Rovo Integration: We didn't just build a sidebar app; we built a true Agent that feels like a team member.

What we learned

  • Rovo is an Orchestrator: We learned that Rovo is most powerful not when it generates text, but when it bridges the gap between unstructured data (photos) and structured workflows (tickets).
  • The Power of Specs: Connecting AI to a "Source of Truth" (Confluence) drastically reduces hallucinations and increases the utility of the agent in professional settings.

What's next for Pit-Stop Inspect Agent

  • Computer Vision Expert: Rovo might call a dedicated Computer Vision model to find the smallest differences (E.g. see unevenly worn tires).
  • Video Analysis: Instead of static images, we want to enable a "Walkaround Mode" where the mechanic takes a video of the car.
  • Inventory Connection: Automatically checking stock levels for replacement parts (e.g., "Front Wing Damaged" -> "Check Inventory for Part #FW-2025").
  • Mobile Optimization: Refining the experience for pit crews using tablets directly on the track.

Further use-cases

Analyzing before-and-after pictures would also be interesting for car rental companies or leasing providers.

Built With

Share this project:

Updates