💡 Inspiration As developers, we’ve all experienced the pain of onboarding into unfamiliar codebases. Important context is hidden in GitHub PRs, Slack threads, and tribal knowledge that vanishes when someone leaves. This frustration sparked an idea:

What if AI could turn our developer activity into usable team knowledge?

We envisioned DevSage as a teammate — one that never forgets why a change was made or how something works, and helps new developers ramp up fast with answers grounded in real team history.

⚙️ What it does DevSage connects to GitHub and Slack, then:

Ingests commits, PRs, and messages

Summarizes project context and decisions

Generates personalized onboarding guides

Answers developer questions using AI + team history

Preserves knowledge across teams, forever

It acts like an internal Stack Overflow — but with your own company’s context.

🛠️ How we built it Bolt.new for rapid full-stack AI application development

Supabase for authentication and secure data storage

GitHub & Slack OAuth for permissions-based data ingestion

OpenAI (via LangChain) for natural language understanding and summarization

Tavus integration (planned) for onboarding via AI-powered video agents

Session-first design with shared Supabase client across all modules

We built a fallback-safe, database-agnostic authentication system so that login never fails — even when Supabase or tokens are unavailable.

🧱 Challenges we ran into GitHub OAuth failed repeatedly due to redirect URI mismatches and strict callback validation

Supabase sessions didn’t persist across clients until we moved to a unified singleton

Bolt’s automatic profile insert logic conflicted with our trigger-based DB structure

Aligning frontend, backend, and Supabase edge cases took heavy debugging

Writing onboarding copy that felt “human” while being AI-generated was tricky

🏆 Accomplishments that we're proud of Created a resilient, production-grade auth flow that never blocks login

Made AI-generated onboarding feel real, not robotic

Implemented trigger-based profile creation and full fallback mode

Built a demo that works both with and without Supabase connected

Deployed a polished GitHub-to-knowledgebase flow within hackathon time

📚 What we learned AI-powered tools are only useful when deeply contextual

Supabase auth requires careful session and redirect management

Developer-focused UX should be invisible — it should just work

A single shared Supabase client solves 90% of auth/session bugs

Hackathon deadlines force smart prioritization of MVPs over polish

🚀 What's next for DevSage 🎥 Integrate Tavus AI video agents to deliver visual onboarding

🧩 Launch VS Code extension to bring DevSage into the developer workflow

🧠 Expand support for Jira, Confluence, Notion for holistic context ingestion

🪄 Introduce chat-based dev assistant that answers architecture questions on Slack

🚀 Onboard early-stage startup teams to refine real-world use cases

Built With

Share this project:

Updates