Forge is an AI-based onboarding platform that turns a production codebase into a sequenced, modular learning experience—like "lego blocks" in an interactive walkthrough. Instead of reading scattered docs or relying on tribal knowledge, developers progress through blocks that explain a system's modules, dependencies, and intent, with contextual file navigation and AI assistance.
Built for the 10,000 AIdeas (AWS) competition.
New engineers often spend weeks onboarding through linear, unstructured documentation and tribal knowledge.
That creates predictable pain:
- Slow time-to-first-contribution
- Heavy dependence on senior engineers
- Knowledge loss when experienced teammates leave
- "Vibe coding" risk from AI-assisted development (shipping without understanding why code works)
Forge addresses this with a pedagogy-first approach:
- Converts a repository into an interactive, step-by-step storyboard
- Explains each block with AI in context
- Links explanations directly to real code
- Ingest a repo
- Point Forge at a Git URL
- Clone and store files
- Reuse an existing parsed build for the same public GitHub commit (so you don't wait twice)
- Parse & structure
- Build a module and dependency graph (AST-style)
- Understand how the codebase is organized
- Generate a storyboard
- Use AI (Amazon Bedrock) to produce 5–10 ordered blocks
- Include a learning objective, explanation, key files/symbols, Mermaid diagrams, and suggested questions
- Keep outputs grounded in the actual repo, not generic advice
- Walk through & explore
- Use a VS Code–style workspace (file explorer, read-only code view, storyboard panel)
- Move through blocks in order (or jump), open linked files, and view diagrams/explanations together
- Chat in context
- Ask questions in a block-scoped chat for each block
- Constrain AI to that block and its dependencies to reduce hallucinations
- Role-based paths
- Choose frontend, backend, infra, or full-stack
- Adapt storyboard order and emphasis to what matters first for that role
- Track progress
- Track completion per block and time-on-block
- Help learners and onboarding owners see progress and potential blockers
Ingest, parse, and storyboard generation run asynchronously:
- You can enter the workspace quickly
- Cloning, parsing, and AI generation continue in the background while you browse files
- The UI shows pipeline state (
cloning → parsing → storyboard → ready) and stays usable throughout
- New hires & junior engineers — A guided mental model of the system and safe, contextual Q&A instead of ad-hoc docs.
- Cross-functional engineers — A curated path that skips irrelevant depth and highlights the interfaces that matter.
- Tech leads & onboarding owners — Reusable onboarding assets and less repetitive mentoring.
- Distributed teams — Asynchronous onboarding that preserves architectural intent and stays in sync with the repo.
- Reduce onboarding time by producing a structured, accurate learning path derived directly from the repository.
- Increase comprehension and confidence via block-scoped AI chat grounded in the current module and its dependencies.
- Support role-based paths (frontend, backend, infra, full-stack) so engineers see what's relevant first.
- Stay cost-conscious for MVP with serverless AWS, caching, and on-demand usage (e.g. Bedrock).
- Frontend: Next.js 16, React 19, TypeScript — VS Code–style layout (file tree, editor, storyboard + chat panel).
- Backend: AWS SAM — API Gateway, Lambda (Node.js 20), DynamoDB, S3, Amazon Bedrock (Nova 2 Lite). Parsing uses tree-sitter; storyboard and chat use Bedrock with constrained context.
For setup, run, and deploy instructions, see backend/README.md and frontend/README.md. Repo layout: frontend/ (Next.js app), backend/ (SAM template + Lambda handlers), and PRD/Forge_PRD.md for the full product requirements document.
Damodar Kamani · Agastya Singh · Krishna Perla · Bowen Xia