Side Hustle Hackathon: Build something with open-source AI at its core that someone would actually pay for. We chose to build DocuFlow AI: an autonomous codebase documentation generator & architecture auditing engine.
Researched the technical pain point: developers & freelancers spend 25-35% of their billable hours deciphering legacy codebases, reverse-engineering routes, and drawing architecture diagrams. Architected a multi-stage pipeline:
1. Open-Source AST Parser (extracts symbols, routes, classes, imports)
2. Dual-Engine LLM Orchestrator (Google Gemini 1M+ context + Local Open-Weights via Ollama / Qwen2.5-Coder)
3. Interactive Web Dashboard with live Mermaid.js diagram rendering and Client Handoff Export.
server/agent/parser.jsCreated codebase AST parser with regex tokenizers for JS/TS/Python/Go/Rustserver/agent/pipeline.jsCreated 5-stage documentation pipeline orchestratorAdd Google Gemini API client, local open-weight model runner, prepackaged demo samples, and an interactive frontend dashboard with Mermaid diagrams.
Implemented Gemini API client with automatic fallback, Ollama connector for air-gapped local inference, Express backend with ZIP upload, and a Tailwind + Lucide + Mermaid.js web application. Tested the CLI runner and verified zero-token heuristic generation.
$ node bin/repodoc.js server/samples/ecommerce-api --provider heuristicSuccessfully generated CODEBASE_DOCUMENTATION.md with 6 files, 170 lines, 9 symbols, and Mermaid component & sequence diagrams.Add UI screenshots to README.md and install DevRelay to save the agent session.
Embedded all 4 product screenshots (Codebase Overview, Architecture & Mermaid diagrams, API & Endpoint Finder, and Code Health Audit) into README.md. Installed and registered DevRelay MCP server and skills.
