🌸 Bloom

One prompt. Full technical blueprint.

Describe any app and Bloom instantly generates a complete, production-ready architecture on an infinite canvas — broken down into layers, functional groups, and concrete implementation tasks.

What It Does

Type something like "create youtube" or "create discord" and Bloom:

1. Identifies the MVP feature scope (no bloat, no missing pieces)
2. Picks the right tech stack for the project
3. Decomposes it into Frontend and Backend layers
4. Breaks each layer into functional groups and concrete implementation tasks
5. Streams the full tree live onto an interactive canvas

Example Output

For "create youtube":

Frontend (React)
├── Video Upload
│   ├── React Dropzone component
│   ├── Upload progress bar
│   └── Video upload API request
├── Video Playback
│   ├── HLS.js player component
│   └── Video playback controls
├── Search
│   ├── Search bar component
│   └── Search results page
└── ...

Backend (Node.js + Express + PostgreSQL)
├── Authentication
│   ├── bcrypt password hash
│   ├── JWT refresh token
│   └── OAuth 2.0 integration
├── Video Management
│   ├── S3 multipart upload
│   ├── HLS transcoding job
│   └── Video metadata storage
└── ...

Setup

Prerequisites: Node.js, Python 3.10+, a Groq API key (free)

git clone https://github.com/your-org/bloom.git
cd bloom
cp .env.example .env
# Add your GROQ_API_KEY to .env

Install dependencies:

npm install
cd services/atomizer && pip install -r requirements.txt

Start services:

# Terminal 1 — Atomizer (AI decomposition engine)
cd services/atomizer && uvicorn main:app --port 8000

# Terminal 2 — Frontend
npm run dev

Open http://localhost:3000

Stack

Service	Tech	Port
Canvas UI	Next.js + React Flow	3000
Atomizer	FastAPI + LiteLLM	8000

Models

Bloom uses Groq for fast, free inference:

• Llama 3.3 70B — best quality (default)
• Llama 3.1 8B — fastest
• Mixtral 8x7B — alternative

.env

GROQ_API_KEY=gsk_...
GROQ_MODEL=groq/llama-3.3-70b-versatile  # optional override