CrewSync

Find and manage duplicate Jira tickets across projects using LLMs and vector embeddings. Full control over how duplicates are handled and resolved. Built with Atlassian Rovo Dev.

Comment

Project Story: CrewSync

The Inspiration In high-performance environments like Formula 1, every millisecond counts. In software engineering, that "drag" often comes from a cluttered, redundant backlog. We noticed a recurring problem in large organizations: the project silo. The Mobile team might be working on a crash report, unaware that the Backend team fixed the same root cause two days ago. Current tools often miss these matches because they only look at single projects or rely on basic keyword matching. We built CrewSync to be the "Pit Crew" for your Jira backlog—an intelligent, cross-project watchdog that stops duplicates before they slow you down.

What it does CrewSync is an AI-powered duplicate detection engine that operates across your entire Atlassian site. It doesn't just look for identical text; it understands intent.

Cross-Project Detection: Breaks down silos by finding duplicates across different teams and projects.

Real-Time Watchdog: Scans in the background as you create or update tickets, flagging collisions before you hit "Save."

Manual Batch Sync: Allows managers to run "Diagnostic Laps" on their backlog to clean up existing technical debt.

Human-in-the-Loop Control: Users can instantly Link, Close, or Ignore flagged duplicates, with the AI learning from every decision.

How we built it We leveraged a cutting-edge Hybrid AI architecture to ensure both speed and accuracy:

Signal Extraction: We use Llama to analyze ticket metadata and generate optimized JQL queries, narrowing down thousands of tickets into a relevant search field.

Semantic Search: Using Amazon Titan, we generate vector embeddings of ticket descriptions. We then apply Cosine Similarity to identify the top three semantic matches—finding duplicates that share meaning, even if they use different words.

Intent Validation: The final candidates are sent back to Llama for a deep-reasoning "Intent Analysis" to confirm the match before alerting the user.

Rovo Dev CLI: The entire backend logic—including complex SigV4 signing and Forge timeout handling—was architected and scaffolded using Atlassian Rovo Dev CLI, significantly accelerating our development velocity. Check the github repo docs/ROVO_JOURNAL.md file for the proof

Challenges we ran into The biggest technical hurdle was handling the Forge platform timeouts when processing large datasets for vector embeddings. We had to optimize our logic to ensure the "Watchdog" felt instantaneous to the user. Additionally, implementing SigV4 authentication for secure communication with AWS services required precise configuration, which we solved by leveraging Rovo’s code generation capabilities to handle the heavy lifting of the security headers.

Accomplishments that we're proud of True Semantic Understanding: We successfully moved beyond "keyword matching." CrewSync can tell that "App crashes on login" and "Authentication failure on mobile" are the same issue.

Cross-Project Scalability: Building a tool that safely and efficiently scans across multiple project boundaries.

Seamless UX: Integrating a complex AI pipeline into a clean, "Williams-inspired" dashboard that feels like a natural part of the Jira experience.

What we learned We learned the power of AI Orchestration. By combining Llama’s reasoning with Titan’s mathematical vector search, we created a system that is far more accurate than using a single model alone. We also discovered how much Rovo Dev CLI can augment a developer's workflow when dealing with complex platform constraints.

What's next for CrewSync Predictive Prevention: Moving from "Duplicate Detection" to "Solution Suggestion"—suggesting the fix for a new ticket based on the resolved duplicate.

Advanced TTL Management: Implementing the front-end slider logic to allow users to manage their embedding cache for cost and performance optimization.

Integration with Bitbucket: Linking code commits from duplicate tickets to consolidate the entire development history into a single "Source of Truth."

Making it Cross Site compatible as well.

Built With

Share this project:

Updates