Inspiration
Engineering teams accumulate technical debt quietly: dead code, duplicate logic, and outdated dependencies slowly reduce development speed. We built The Optimizer to make code health visible and actionable before it turns into larger delivery risk.
What it does
The Optimizer analyzes a repository for high-confidence code health issues, including dead code, duplicated logic, unused dependencies, and obvious code smells. It then creates clear GitLab issues with context and recommendations so teams can fix problems quickly.
How we built it
We implemented The Optimizer as a custom flow on the GitLab Duo Agent Platform. The flow uses an agent with repository exploration and analysis tools (list_dir, find_files, read_file, grep) and GitLab issue automation (create_issue). We tuned the system prompt for conservative, high-signal findings and issue output quality.
Challenges we ran into
Balancing precision and recall was the hardest problem. Overly aggressive detection can create noisy issues, while conservative detection can miss useful findings. We also had to design prompts that work across varied codebase structures and languages without introducing too many false positives.
Accomplishments that we're proud of
We built an end-to-end flow that not only identifies technical debt patterns, but also turns them into actionable work items in GitLab. We are especially proud of the practical focus: every finding is framed to help teams decide, prioritize, and fix.
What we learned
We learned that issue quality matters more than issue volume. Teams trust automation when findings are specific, explainable, and reproducible. We also learned that clear agent constraints and structured output requirements significantly improve reliability.
What's next for The Optimizer
Next, we plan to add severity scoring, language-specific analyzers, trend reporting over time, and smarter deduplication of findings. We also want to support remediation suggestions and optional auto-generated merge requests for safe, low-risk fixes.
Built With
- github
- gitlab

Log in or sign up for Devpost to join the conversation.