StackShadow – Autonomous AI CTO Co-Pilot
A multi-agent AI system that connects to GitHub to autonomously map dependencies, detect vulnerabilities, and optimize infrastructure spend.
Project Overview
StackShadow is an autonomous "CTO Co-Pilot" built during the Google Build with AI Hackathon (where it won 2nd place). It utilizes a custom multi-agent architecture powered by the Gemini API and GitHub APIs to automatically audit codebases. The agents work in tandem to map project dependencies, identify security vulnerabilities, and analyze infrastructure usage to recommend actionable cost-saving measures, all surfaced through a Next.js and Supabase backend.
Problem It Solves
Engineering teams often lose visibility into their infrastructure stack as projects scale, leading to bloated cloud bills, unnoticed security vulnerabilities, and outdated dependencies. Manually auditing an entire codebase's architecture and cost footprint is time-consuming and prone to human oversight.
Tools & Technologies Used
Skills Involved
Challenges
The core challenge was coordinating multiple AI agents to work reliably without drifting from their objectives (Goal Hijacking). I had to implement precise LLM function calling to ensure the agents could accurately fetch, parse, and reason over real-time data from GitHub APIs without hallucinating dependencies or costs.
Learnings
This project solidified my transition into specialized AI Engineering. I learned how to move beyond basic LLM wrappers to build robust, autonomous agent workflows. It gave me hands-on experience in managing agentic state, enforcing strict tool boundaries, and bridging complex external APIs with AI-driven logic.