
CodeOracle: AI-Powered Code Intelligence Platform CodeOracle is an architectural decision support engine designed to help software engineers make confident decisions about unfamiliar codebases through multi-agent AI analysis. Built for the IBM watsonx.ai Hackathon, it addresses a fundamental problem in software engineering: engineers are often asked to make changes RIGHT NOW to codebases they don't understand. Traditional solutions like documentation, onboarding, or manual code review take too long and don't provide the real-time risk assessment and impact analysis needed under uncertainty. CodeOracle solves this by combining repository ingestion, multi-agent AI analysis powered by IBM watsonx.ai Granite models, a FastAPI backend, and an intuitive React frontend to deliver actionable insights in seconds. The entire workflow, from uploading a repository to receiving a complete risk analysis with impact predictions, takes under 90 seconds, making it ideal for the hackathon's "wow moment" requirement. Architecture: 1. Repository Ingestion Pipeline 2. Multi-Agent AI Architecture Agent 1: RepoMapper Agent 2: DependencyAnalyst Agent 3: RiskDetector Agent 4: KnowledgeSynthesizer Agent 5: ImpactSimulator Supervisor Agent The system supports three LLM providers through a common abstract interface: Groq API for fast free-tier access using Llama models, Ollama for fully local operation with no rate limits or data leaving the machine, and IBM watsonx.ai for the target Granite model integration required by the hackathon. Each provider implements JSON response parsing with fallback extraction from markdown code blocks, and the Groq provider includes exponential backoff retry logic for handling rate limits.
17 May 2026