Qodo: AI-Powered Multi-Agent Code Integrity Platform
Qodo, an NVIDIA Inception program member, has developed a multi-agent code integrity platform that enhances and automates software quality workflows with AI-powered agents for code writing, testing, and review. Qodo's core is its deeply context-aware AI agents, built on advanced retrieval-augmented generation (RAG) and a state-of-the-art code embedding model trained on NVIDIA DGX. This allows the agents to understand code intent, patterns, and broader structure, generating more accurate code suggestions, reliable tests, and insightful code reviews. To achieve this, Qodo built a robust pipeline for continuously maintaining an up-to-date index and employs language-specific static analysis for chunking large code files, avoiding issues caused by invalid or incomplete code segments impacting LLM performance. In a collaboration with NVIDIA, Qodo's solution significantly improved the accuracy of NVIDIA's internal RAG system (Genie) for code search, ultimately integrated into NVIDIA's internal Slack system for providing developers with more accurate and detailed code information.