Built an autonomous, Docker-sandboxed DevOps agent that automatically detects test failures, analyzes full codebase context using Google's Gemini LLM, and generates precise patches to self-heal broken code. Key highlights: Designed a self-healing loop (run tests → send failure logs to Gemini → apply patch → retest) that iterates until all tests pass or a configurable retry limit is hit Implemented Docker sandboxing (via dockerode) to safely execute test suites in isolated containers, protecting the host environment from untrusted code execution Integrated Google Gemini API with structured JSON output (responseSchema) for reliable, parseable fix suggestions Built an AI-powered code generation playground where users type natural language prompts (e.g. "write a program to print prime numbers") and instantly get generated code with explanations, across multiple languages (JavaScript, Python, C++, Java, Go, Rust, HTML, CSS) — including live preview for HTML/CSS and output execution for others Built a real-time monitoring dashboard using Server-Sent Events (SSE) to stream live execution logs and code diffs to the frontend Added a file backup and rollback system so every applied fix is reversible if it introduces regressions.