Expert knowledge, forged into tools for everyone.
Forge captures what experts know through natural conversation and transforms it into interactive tools anyone can use. No forms. No manuals. Just speak.
Built for the Claude Code Hackathon - Break the Barriers problem statement.
- Speak Your Knowledge - An AI interviewer conducts a structured conversation, asking the right questions to draw out what you know. Voice or text.
- Knowledge Extraction - As you talk, Opus extracts structured knowledge in real-time: facts, processes, decisions, pitfalls.
- Interactive Tool - Your knowledge becomes a living tool: decision trees, checklists, calculators, quizzes. Share a link and anyone benefits.
Opus 4.6 isn't just the model - it's the architect. Seven distinct roles orchestrate the pipeline:
| Role | What It Does |
|---|---|
| Interview Planner | Designs structured interview sections with extraction priorities based on domain analysis |
| Conductor | Manages conversation flow in real-time, deciding when to probe deeper vs advance |
| Knowledge Extractor | Pulls structured facts, processes, and decisions from natural conversation |
| Tool Architect | Selects the right component types and designs tool layout from extracted knowledge |
| Expert Channel | Answers user questions by channelling expert knowledge through 5-layer cascading context |
| Tool Refiner | Understands conversational edit requests and updates component configs in real-time |
| Knowledge Integrator | Analyzes follow-up interviews and proposes updates to existing tools |
- Decision Trees - Branching logic with recommendations
- Checklists - Requirements and readiness validation
- Step-by-Step Guides - Sequential procedures with tips
- Calculators - Quantitative assessments with formulas
- Question Flows - Intake questionnaires with AI advice
- Quizzes - Knowledge checks and scenario testing
- Score Cards - Multi-criteria evaluation
- Comparison Tables - Side-by-side analysis
- Info Cards - Key reference information
- Risk Assessments - Risk identification and mitigation
- Context Panels - Situational guidance
| Layer | Technology |
|---|---|
| Frontend | React, TypeScript, Tailwind CSS, Vite |
| Backend | Bun, Hono, Server-Sent Events |
| Database | PostgreSQL 16, Drizzle ORM, pgvector |
| AI Models | Claude Opus 4.6, Claude Sonnet 4.5 |
| Voice | ElevenLabs Conversational AI |
| Testing | Playwright (31 E2E tests) |
# Prerequisites: Docker, Bun, Node.js 20+
# Clone and install
git clone https://github.com/bigadamknight/forge.git
cd forge
yarn install
# Start PostgreSQL
docker compose up -d
# Configure environment
cp .env.example .env
# Add your ANTHROPIC_API_KEY and ELEVENLABS_API_KEY
# Run migrations and seed data
./stack db:migrate
yarn workspace @forge/db seed
# Start the app
./stack start
# Frontend: http://localhost:3070
# API: http://localhost:3071apps/
api/ Bun + Hono REST API with SSE streaming
web/ React SPA (Vite + Tailwind)
packages/
db/ Drizzle schema, migrations, seeds
shared/ Shared TypeScript types
mcp/ MCP server for forge knowledge
e2e/ Playwright E2E tests
./stack api # API server on :3071
./stack web # Frontend on :3070
./stack typecheck # TypeScript checks
./stack db:migrate # Run migrations
./stack db:psql # Database shell
# E2E tests (requires running app)
yarn test:e2eAdam Knight - @bigadamknight
Built entirely with Claude Code (Opus 4.6).
0 comments
log in to comment.