Codebase

Inspiration

Every developer has that one folder — old projects full of logic they once mastered but can’t recall now. We realized: code doesn’t forget, but developers do. So we built Codebase.AI — a platform that keeps your resting code alive and helps new developers learn faster by chatting directly with the knowledge inside past projects.

What it does

Codebase.AI turns your code into a living, searchable memory. Upload your repositories or connect GitHub, and instantly start chatting with your codebase. You can ask things like: “Where did we handle login validation before?” “Show me all UPI transaction functions across our apps.” “Explain how this Vue component works — like I’m new to the team.” It’s both a memory engine for senior developers and a learning companion for newcomers, turning years of accumulated logic into conversational, understandable insights.

How we built it

Frontend: Built with React + Next.js for a dynamic AI chat experience. AI Layer: Powered by Vertex AI for summarization, reasoning, and teaching explanations. Vector Engine: Elastic Cloud with text-embedding-004 to index and retrieve semantically similar code. Backend: Node.js serverless stack managing embeddings, uploads, and chat context. Authentication: Individual & Organization tiers (80% Done, so in progress) with roles (Admin, Contributor, Viewer). When a repository is uploaded, it’s automatically chunked, embedded, and indexed in Elastic. Gemini then uses those embeddings to generate context-rich, educational responses that help both experienced and new developers navigate and reuse code.

Challenges we ran into

Challenges we ran into Designing chunking logic that respects file structure and function boundaries. Preserving semantic relationships between comments, code, and documentation. Keeping latency low while serving responses from large repositories. Teaching the AI to adapt tone — from technical precision to onboarding friendliness.

Accomplishments that we're proud of

Built a system that truly remembers what developers built before — down to function logic. Created a chat experience that doubles as a learning mentor for onboarding developers. Made cross-project code reuse as simple as asking a question. Blended Vertex AI Gemini reasoning with Elastic’s precision for fast, meaningful results.

What we learned

AI can bridge the gap between “I once built this” and “I understand this again.” Elastic embeddings combined with Gemini reasoning can revive dormant codebases with near-human comprehension. Developers value not just answers — but explanations that teach, like pair-programming with your past self.

What's next for Codebase

Work on finishing organisation mode (80% there) Introduce Org Learning Mode, where new members can learn from every past project automatically. Launch “Code Stories”, turning major functions or modules into readable tutorials.

Built With

• elasticsearch
• gcp
• gemini
• javascript
• node.js
• vertex