Skip to content

Latest commit

 

History

9 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

VentureMind AI

Your startup idea. Six AI analysts. A BUILD / CAUTION / REJECT verdict in 90 seconds.

Founders don't need another idea validator that tells them their idea is "promising" — that's what every AI tool says. VentureMind runs your idea through six specialist agents that argue like an actual investment committee, then forces a deterministic score out the other end. No hedging. No "it depends." A verdict.


The Problem

Most idea-validation tools are a single LLM call wearing a nice UI — ask a chatbot "is this a good idea?" and it will find a way to say yes. Founders walk away validated and unprepared, because the tool never actually modeled how an investor would push back.

VentureMind is built to disagree with you. It runs live market research, maps real competitors, and only reaches a verdict after a quality-gate agent has checked whether the evidence actually supports the conclusion — reflecting and re-researching if it doesn't.


How It Thinks: The Six-Agent Committee

# Agent Role
01 Planner Breaks the idea into concrete research vectors — what actually needs to be checked
02 Market Searches the live web for real demand and timing signals
03 Competitor Maps the actual competitive landscape, not a generic list
04 Evaluator The quality gate — if evidence is weak, it triggers a reflection loop and sends agents back to dig deeper
05 Decision Deterministic scoring: demand × 0.5 + competition × 0.3 + risk × 0.2 — same inputs always produce the same verdict
06 Report Synthesizes SWOT, key findings, and a full investor memo

Why the Evaluator matters: most "multi-agent" hackathon projects are really just several prompts chained in a fixed sequence — call A, feed to B, feed to C, done. VentureMind's pipeline isn't fixed-length: the Evaluator can reject its own committee's early findings and send research agents back for another pass before a verdict is allowed to form. That's a real feedback loop, not a relay race.

Why deterministic scoring matters: ask most AI tools the same startup idea twice and you'll get two different opinions with two different confidence levels. VentureMind's Decision agent applies a fixed formula on top of the researched inputs — the verdict is reproducible, defensible, and auditable, which is what an actual investment committee needs and what a "vibe check" tool can't offer.


Architecture — One Product, Two Services

┌─────────────────────────────────────────────────┐
│  Next.js 14 Frontend  (Vercel)                   │
│  Premium SaaS UI · shadcn/ui · TypeScript        │
│  → POST /api/analyze (proxied to backend)        │
└─────────────────────┬───────────────────────────┘
                      │ HTTP JSON
┌─────────────────────▼───────────────────────────┐
│  FastAPI Backend  (Render / Railway / VPS)       │
│  api_server.py → POST /analyze → ResearchRunResult│
│  6 Python agents · Groq LLM · Serper live search │
└───────────────────────────────────────────────────┘

No backend? No problem. If NEXT_PUBLIC_API_URL isn't set, the frontend runs in demo mode — the full UI, with illustrative analysis, so anyone can experience the product with zero setup and zero API keys. Judges never see a broken build.

A standalone Streamlit UI (app.py) ships alongside the Next.js frontend for anyone who wants the Python-only experience with no Node.js at all.


Quick Start — Local Full Stack

1. Clone and install

git clone https://github.com/code-withkrishna/VentureMind-AI
cd VentureMind-AI

pip install -r requirements.txt   # Python backend
npm install                      # Next.js frontend

2. Set API keys

cp api.env.example api.env
# add GROQ_API_KEY and SERPER_API_KEY

3. Start the backend

uvicorn api_server:app --host 0.0.0.0 --port 8000 --reload
# → http://localhost:8000  (docs at /docs)

4. Start the frontend

NEXT_PUBLIC_API_URL=http://localhost:8000/analyze npm run dev
# → http://localhost:3000

Or, Python-only in 5 minutes, no Node.js:

pip install -r requirements.txt
cp api.env.example api.env
streamlit run app.py
# → http://localhost:8501

Production Deployment

Backend → Render (free tier)

  1. render.com → New Web Service → connect this repo
  2. Build command: pip install -r requirements.txt
  3. Start command: uvicorn api_server:app --host 0.0.0.0 --port $PORT
  4. Health check path: /health
  5. Environment variables: GROQ_API_KEY, SERPER_API_KEY, FRONTEND_URL, ENVIRONMENT=production

Frontend → Vercel (free tier)

  1. vercel.com → New Project → import repo (Next.js auto-detected)
  2. Environment variable: NEXT_PUBLIC_API_URL=<your-render-url>/analyze
  3. Deploy.

Streamlit-only deployment

  1. Push to GitHub
  2. share.streamlit.io → New app → entry point app.py
  3. Add secrets: GROQ_API_KEY, SERPER_API_KEY

Project Structure

├── api_server.py          FastAPI server — entry point for the Next.js frontend
├── app.py                 Standalone Streamlit UI (no Node.js needed)
├── main.py                CLI runner
├── requirements.txt       Python dependencies (fastapi + uvicorn included)
├── agents/                The 6 specialist agents
├── core/                  Orchestrator, models, config, providers
├── memory/                SQLite run history
├── tools/                 Search and calculator tools
├── ui/                    Streamlit rendering module
├── utils/                 PDF generator, scenario engine
└── src/                   Next.js 14 frontend
    ├── app/
    │   ├── api/analyze/route.ts   Proxies to ANALYZE_API_URL
    │   └── page.tsx               Main UI + demo mode fallback
    ├── components/
    │   ├── ui/                    shadcn/ui primitives
    │   └── venturemind/           Product components
    └── lib/utils.ts               Mock generator + utilities

Environment Variables

Variable Required Description
GROQ_API_KEY ✅ Python backend Groq LLM key
SERPER_API_KEY ✅ Python backend Live search key
NEXT_PUBLIC_API_URL Next.js only URL of the api_server.py analyze endpoint
ANALYZE_API_KEY Optional Bearer token if the backend is auth-protected

What's Next

  • Scenario studio: run the same idea through 5 investor-persona lenses (already scaffolded via the scenario engine in utils/) and surface it in the main UI, not just the PDF export
  • Expose the reflection-loop trace in the UI — show which agent rejected the first pass and why, turning the quality gate from a backend detail into a visible trust signal
  • Persist run history (already stored in memory/) as a comparable dashboard, so a founder can re-run an idea after a pivot and see the verdict shift

VentureMind is decision support for early-stage ideas — a fast, structured second opinion before you spend months and money finding out the hard way.

About

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages