Code-Eagle AI
End-to-end pull-request review system using Qwen-Coder, QLoRA, Tree-sitter AST context, and static analysis.
Problem
Manual PR review is slow and inconsistent, and generic LLM reviewers lack the structural context of the codebase.
Approach
Fine-tuned Qwen-Coder with QLoRA; Tree-sitter AST context and static analysis feed the reviewer; GitHub webhooks, RabbitMQ, and FastAPI microservices run the pipeline.
Progress
Work in progress. 40% less manual review time in current trials.