RepoRelic

Inspiration

As a solo developer, manual debugging felt like endless detective work—figuring out where code might "crack" was troublesome. I wanted automated analysis that learns from code history to make debugging smarter and faster.

What it does

Automates the Manual:

Generates tests automatically instead of writing them by hand Identifies risky functions without code review Runs all tests and diagnoses failures in one command Instant Insights:

Single report shows complexity, issues, dependency problems, and refactoring opportunities No hunting through linters, test results, or documentation Actionable suggestions (remove unused code, consolidate duplicates, fix risky patterns) Solo Dev Multiplier:

What takes a team hours to audit (code quality, test coverage, dependencies) takes seconds Reduces context-switching—everything needed for a review is in one report Catches bugs before deployment instead of in production Bottom Line: Replaces manual testing, code review, and analysis with one click. Saves 2-4 hours per analysis cycle.

How we built it

Ideation: Balancing automation with actionable insights—early test generation was noisy, requiring graph refinements.

Marketplace Deployment: Azure DevOps pipelines failed repeatedly due to authentication and LICENSE issues for the extension.

Technical: Multi-provider API handling, performance scaling for large codebases, and cross-platform compatibility.

Challenges we ran into

RepoRelic is a VS Code extension with a Python engine: Extension (TypeScript): UI, commands, and settings for LLM providers. Engine (8 Stages): Parse code → static analysis → dependency graphs → knowledge enrichment (with git history) → LLM test generation → execution → diagnosis → Markdown reports.

LLM Client: Unified interface supporting multiple providers with rate limiting.

Accomplishments that we're proud of

Git Mining: Historical bug patterns from commits improved test generation and reliability scoring.

Incremental Design: Modular 8-stage pipelines enhance maintainability over monolithic scripts.

What we learned

LLM Integration: Balancing multiple AI providers (OpenAI, Gemini, DeepSeek) taught API management and prompt optimization, where cost scales with tokens. Graph Theory,

What's next for RepoRelic

Support for more languages. Adding a feature to consider previous git versions and leverage knowledge graph to see which merges cause the specific issue helping in root cause analysis

Built With

• google-genai
• jinja
• networkx
• openai
• pylint
• pytest
• python
• radon
• typescript