Built a code review pipeline on top of qwen2.5-coder — runs locally, zero code sent anywhere, finds AI-generated code bugs

Article automatically generated from technical news.

Been running qwen2.5-coder locally for code review and the results are genuinely useful. Built a full pipeline around it called DevScan AI. What it does: - Fetches code from any public GitHub URL - Runs Bandit SAST for security scanning - Sends to local Ollama for deep AI review - Returns security score, quality score, bug list Temperature=0 throughout. Same code always gives same result. No randomness. ━━━━━━━━━━━━━━━━━━━━ MODEL PERFORMANCE ━━━━━━━━━━━━━━