PitCrew AI for Atlassian

dashboard screen

🏁 Inspiration

Modern software teams move fast, but their pull request workflows often lack real-time visibility and prioritization.

We were inspired by how Formula 1 pit crews operate under pressure: they rely on live telemetry, clear signals, and rapid decision-making to prevent failures before they happen.

In contrast, development teams often face:

Risky pull requests merged too late
Critical changes hidden among dozens of low-risk PRs
Reviewers overloaded without clear prioritization
Bottlenecks discovered only after delays occur

We wanted to bring the clarity, anticipation, and operational discipline of motorsport telemetry into Atlassian tools.

That idea became PitCrew AI for Atlassian.

🧠 What it does

PitCrew AI turns Bitbucket and Jira into a real-time engineering telemetry system that helps teams detect risk early and focus their attention where it matters most.

Core capabilities:

Analyzes every pull request in real time as soon as it is created or updated
Calculates an explainable risk score (0–100) based on:
- Changed files and their criticality
- Lines added and removed
- Review context and process signals
Classifies PRs into clear risk levels (green, yellow, red)
Posts automatic, non-spammy comments in Bitbucket with a full risk breakdown
Displays live telemetry dashboards showing PR health, risk distribution, and team load
Flags high-risk PRs early, before they slow down or destabilize a sprint

The objective is simple: Help teams see problems earlier, review smarter, and keep delivery predictable.

🛠️ How we built it

PitCrew AI is built entirely on Atlassian Forge, using a serverless and Atlassian-native architecture.

Backend and analysis:

Bitbucket webhooks trigger PR analysis on creation and updates
A custom risk scoring engine evaluates:
- File types and critical paths
- Change size and structure
- Process signals like reviewers and timing
Smart gating and caching prevent unnecessary re-analysis
All webhook payloads are validated with runtime schemas
Results are stored using indexed Forge storage for fast dashboard queries

Bitbucket integration:

A single persistent comment per PR is created or updated automatically
Comments include:
- Visual risk indicators
- Key metrics
- A transparent explanation of every scoring factor

Dashboard:

A React-based dashboard embedded in Jira
Multiple telemetry views inspired by race control screens
Real-time data sourced directly from Forge storage
Designed for fast scanning and decision-making

The system is optimized for speed, robustness, and clarity, with no external servers and no access to source code.

🚧 Challenges we ran into

Designing a risk model that is fast and explainable, without relying on heavy machine learning
Avoiding noisy automation while still providing immediate feedback
Handling Bitbucket events reliably at scale with retries, caching, and idempotency
Keeping the motorsport-inspired UI expressive without becoming gimmicky
Ensuring security, least-privilege access, and data minimization from day one

Balancing usefulness, transparency, and performance was the main technical challenge.

🏆 Accomplishments that we're proud of

A production-grade PR risk analysis pipeline running entirely on Forge
An explainable scoring system that developers can actually trust
Automatic Bitbucket comments that update intelligently without spam
A visually distinctive yet practical telemetry dashboard
A clean architecture with structured logging, validation, and security documentation

Most importantly, we built a tool that feels immediately useful in real development workflows.