Inspiration

We saw how developers spend nearly 50% of their time debugging—often getting bogged down in repetitive tasks instead of building new features. Inspired by projects like ChatDBG—which uses LLMs to enable conversational debugging through GDB/PDB—and Dynatrace’s Live Debugger, we wanted to create a tool that did more than just highlight errors: a real-time, IDE-integrated assistant that not only spots bugs but explains, fixes, and tests them automatically

 What it does

Live error detection: Continuously monitors your code as you type. Root-cause analysis: Explains why the bug occurs using context-aware reasoning. Auto-fixing: Generates code patches or suggestions automatically. Test generation: Builds unit-test scaffolding or regression tests to validate fixes. IDE integration: Works seamlessly in VS Code (and beyond) via an extension.

we built it

AST + LLM hybrid engine: Parsed code with an AST parser, fed critical snippets into GeminiAI for diagnosis Context management: Captured code history and runtime snapshots to give the AI runtime context. IDE extension: Created a VS Code plugin to surface suggestions inline and apply quick-fix actions. Test scaffolding: Called GeminiAI to generate test templates and file changes.

🧩 Challenges we ran into Performance tuning: Balancing depth of analysis with latency to keep suggestions under 500ms .

Context window limits: Truncating long files while preserving enough context for LLM understanding.

Fix safety: Generating reliable fixes without introducing regressions or new bugs.

 Accomplishments we're proud of

50 developers user-tested, with feedback showing ~60% of auto-fixes rejected—but ~30% accepted with minimal edits. Explains root cause correctly ~67% of the time on curated bug set. Seamless "one-click fix + test" flow demoed live in IDE with zero config.

 What we learned

Keep scope tight: Simple, real-world utility (“find–fix–test flow”) resonates more than broad feature sets Context is key: Even powerful LLMs need AST parsing and runtime info to diagnose complex issues. Iterative UX matters: Real-time feedback is critical—waiting 5+ seconds kills developer trust.

What’s next for DebugMate

Multilanguage support: Expand beyond Python to JS, Java, Go. Time-travel debugging: Add “rewind and step” feature akin to Cursor or Docombug. CI/CD integration: Run tests and auto-fixes in pipelines (GitHub Actions, GitLab). Community model: Make it open-source, support plugin architecture—and cultivate contributor base.

Built With

• api
• boltai
• css
• geminiai
• git
• github
• html
• netlify
• react
• tailwind
• typescript
• visual-studio
• vite