From raw idea to the single smallest testable action in 4 steps. An agentic AI pipeline that structures startup ideas, extracts riskiest assumptions, generates a human-approved 30/60/90-day roadmap, and isolates your very first step.
Built for the USAII Global AI Hackathon 2026
Undergraduate Track · Challenge Brief 3: Build the "Second Brain" for Real Life · Direction B: Zero-to-One Builder.
Zero-to-one builders face cognitive overload. You have a raw idea — maybe just a sentence — but translating it into a structured plan feels overwhelming. What are your riskiest assumptions? What should you build first? What's the smallest thing you can do today to get real signal?
Most people never start. Not because the idea is bad, but because the gap between "I have an idea" and "here's what I do next" is paralyzing.
IdeaForge bridges that gap with four specialized AI agents that structure your thinking, surface your blind spots, plan your path, and break the paralysis — all while keeping you in control.
| Stage | Agent | What It Does |
|---|---|---|
| 01 · Intake | 6W3H Chatbot | Structures your raw idea across 9 dimensions (Who/What/Where/When/Why/Which/How/How much/How many). AI infers gaps but never overwrites your input. |
| 02 · Assumptions | Risk Analyst | Surfaces the 3–5 riskiest unvalidated beliefs about your idea (demand, willingness to pay, distribution). Each tagged low/medium/high. |
| 03 · Roadmap | Execution Coach | Synthesizes a 30/60/90-day milestone plan. Human-in-the-Loop: every milestone stays draft until you explicitly accept it. |
| 04 · First Step | Paralysis Breaker | Isolates the single smallest testable action — 24 hours, no budget, no team — that produces real signal about your riskiest assumption. |
IdeaForge is structured as a commercial-grade full-stack application with completely decoupled frontend and backend services:
- Frontend (
/frontend): Next.js App Router (React 19) with Tailwind CSS 4, shadcn/ui, Framer Motion, and Zustand - Backend (
/backend): Dedicated Express.js server exposing REST API endpoints - Database: SQLite with Prisma ORM (stored in
/backend/prisma) - AI Layer: 4 specialized agents utilizing
@google/generative-ai(Gemini API)
📐 See ARCHITECTURE.md for the full system diagram, data flow, and design decisions.
We take AI responsibility seriously. Our design addresses five risk categories with concrete, code-level mitigations:
| Risk | Mitigation |
|---|---|
| Over-reliance | HITL milestone acceptance — AI proposes, human disposes |
| Algorithmic bias | Transparent risk tags + diverse prompt engineering |
| Content safety | Regex moderation layer blocks harmful input pre-AI |
| Hallucination | Structured JSON contracts + bounded agent scope |
| Data privacy | Local SQLite, no PII, session-based identity |
📋 See RESPONSIBLE_AI.md for the full risk analysis and mitigation documentation.
The project is structured as a monorepo with a single root configuration to launch both the frontend and the backend simultaneously.
# 1. Install all dependencies (Root, Frontend, Backend)
npm run install:all
# 2. Start the development servers
npm run devThe application will start automatically:
- Frontend runs on
http://localhost:3000 - Backend API runs on
http://localhost:3001
ideaforge/
├── frontend/ # Next.js Application
│ ├── src/
│ │ ├── app/ # App Router pages & layouts
│ │ ├── components/ # UI components (shadcn, chatbot, views)
│ │ ├── lib/ # API clients, types, and Zustand store
│ │ └── hooks/ # Custom React hooks
│ ├── public/ # Static assets
│ ├── next.config.ts # Next.js configuration
│ └── package.json # Frontend dependencies
│
├── backend/ # Express.js REST API
│ ├── src/
│ │ ├── index.ts # Express server entry point & routes
│ │ ├── config/ # Database and environment config
│ │ ├── schemas/ # Zod validation schemas
│ │ ├── services/ # The 4 AI Agents implementation
│ │ └── geminiClient.ts # Gemini API integration
│ ├── prisma/ # SQLite Database & ORM Schema
│ │ ├── schema.prisma # Data models (Roadmap, Assumption, etc)
│ │ └── dev.db # Local SQLite database
│ └── package.json # Backend dependencies
│
├── package.json # Root orchestrator (concurrently)
├── README.md # You are here
├── ARCHITECTURE.md # Technical documentation
└── RESPONSIBLE_AI.md # Ethical AI documentation
| Method | Path | Purpose |
|---|---|---|
| GET | /api/health |
Server + database health check |
| GET | /api/session |
Session management (anonymous UUID) |
| GET | /api/roadmaps |
List user's roadmaps |
| POST | /api/intake |
Create roadmap from raw idea + 6W3H AI Agent |
| GET | /api/intake/:roadmapId |
Fetch a specific roadmap |
| POST | /api/assumptions/:roadmapId/generate |
Run assumption extraction AI Agent |
| GET | /api/assumptions/:roadmapId |
List assumptions for a roadmap |
| POST | /api/milestones/:roadmapId/generate |
Run milestone generation AI Agent |
| GET | /api/milestones/:roadmapId |
List milestones for a roadmap |
| PATCH | /api/milestones/:milestoneId/accept |
HITL accept a milestone |
| POST | /api/first-step/:roadmapId/recommend |
Run first-step recommendation AI Agent |
| GET | /api/first-step/:roadmapId |
Fetch first-step recommendation |
Proudly engineered and designed for the USAII® Global AI Hackathon 2026.
| Team Member | Role | Focus Area |
|---|---|---|
| Talha Abid | Frontend Developer | React / Next.js / Tailwind CSS (Frontend, QA, & Deployment) |
| Eman Mirza | Backend Developer | Node.js / Express Architecture (Backend, QA & Deployment) |
| Mahad | Database Engineer | Data Modeling & Prisma ORM |
| Mahrukh | Product Strategy | Brainstorming & Problem Solving |
MIT License
Forged for the ambitious · USAII Global AI Hackathon 2026
Undergraduate Track · Challenge Brief 03 · Direction B: Zero-to-One Builder