SafeSeedOps Lite

Live Demo: seed-ops-lite.onrender.com

SafeSeedOps Lite is an enterprise-grade synthetic relational database generator using a multi-agent architecture — winner of the AMD Developer Hackathon 2026 Track 3 (Unicorn). This project provides the backend foundation, schema visualizer, and relational synthesis suites with native AMD ROCm support.

---

Quick Start

1. Environment Configuration

Copy the example environment file and set at least your Fireworks AI API key:

cp .env.example .env

Edit .env and set FIREWORKS_API_KEY=your_key_here. See the Deployment section for all supported environment variables.

2. Developer Environment Bootstrap

Prerequisites: Ensure Python 3.10+ and uv are installed:

pip install uv

If uv is not recognized after installing (Windows), add the Python Scripts folder to your PATH:

$scripts = "$(python -m site --user-base)\Python310\Scripts"
[Environment]::SetEnvironmentVariable("Path", "$env:Path;$scripts", "User")

Then restart your terminal, or run $env:Path += ";$scripts" in the current session.

Install project dependencies:

uv sync

3. Unified Developer Startup

Launch both frontend and backend development environments automatically:

uv run seed dev

4. Advanced Alternative (Manual Startup)

If you prefer running the processes in separate terminal instances manually:

• Start the Backend:

  uvicorn app.main:app --reload --port 8000

• Start the Frontend:

  cd frontend
  npm install
  npm run dev

• Run the Demo Wizard:

  uv run python scripts/demo_wizard.py

---

AMD Integration

SafeSeedOps Lite was built for and won the AMD Developer Hackathon 2026 Track 3 (Unicorn). It supports three LLM backends auto-routed in priority order:

1. ROCm + Gemma (local) — fastest, zero API cost
2. Fireworks AI (cloud) — Mixtral 8x7B for high-throughput
3. Fallback — Gemini / OpenAI / Anthropic

Fireworks AI (Cloud)

Set your API key in .env:

FIREWORKS_API_KEY=your_key_here

The default cloud provider uses Mixtral 8x7B via Fireworks AI for scalable generation.

ROCm Local Inference

Requires an AMD GPU on Linux with ROCm 5.7+ installed. Auto-detected via rocm-smi at startup — if the command is available and returns GPU data, the local inference path is enabled automatically.

Download Gemma models for local inference:

uv run python scripts/download_model.py download gemma-2-2b-it

Supported local models:

Model	Parameters	Min VRAM
Gemma 2 2B	2.5B	4 GB
Gemma 2 9B	9B	8 GB
Gemma 3 12B	12B	12 GB
Llama 3.2 3B	3.2B	4 GB

Auto-Routing Logic

At generation time, the LLM Gateway selects the best available provider:

Gemma models requested  →  ROCm (if available)
                              ↓
                           Fireworks AI (if key set)
                              ↓
                           Fallback provider

---

Key Features

• Interactive Schema Designer: Build and configure schemas locally.
• PostgreSQL DDL Import: Parse and convert SQL scripts into structured relational definitions.
• Cost-Aware Topological Planner: Computes sequence validation gates and execution plans for database generation.
• Diagnostics Health Panel: System pre-flight warnings and fallback checks.
• Multi-Provider LLM Support: Fireworks AI, ROCm + Gemma, Gemini, OpenAI, and Anthropic.
• AMD ROCm Integration: Native local Gemma inference on AMD GPUs via ROCm stack.
• Interactive Demo Wizard: uv run python scripts/demo_wizard.py for a guided generation walkthrough.

---

Architecture

SafeSeedOps Lite uses a multi-agent pipeline where each stage is independently scalable:

graph LR
    A[Schema Designer] --> B[Validation Engine]
    B --> C[Topological Planner]
    C --> D{LLM Gateway}

    D --> E[Fireworks AI]
    D --> F[ROCm + Gemma]
    D --> G[Gemini]

    E --> H[Data Generator]
    F --> H
    G --> H
    H --> I[Export: JSON / CSV / Parquet]

Loading

1. Schema Designer — Parse DDL or use the visual schema builder
2. Validation Engine — Check column types, PK/FK constraints, nullability
3. Topological Planner — Compute table dependency DAG (parents before children)
4. LLM Gateway — Route to the best available provider (ROCm → Fireworks → Fallback)
5. Data Generator — PK-first, relationship-aware row generation with streaming
6. Export — JSON, CSV, or Parquet output

---

Project Structure

• app/: FastAPI Backend routing, validation, planning, and generation services.
• frontend/: React SPA user interface.
• tests/: Verification suites and performance benchmark scripts.
• docs/: Design documents and technical manuals.

---

Testing & Quality Check

# Format verification
black app/ tests/

# Linter checks
ruff check app/ tests/

# Strict type checks
mypy app/

# Unit & integration tests
pytest

# AMD end-to-end validation
uv run python scripts/e2e_amd_check.py

# Provider benchmarks
uv run python scripts/bench_providers.py

---

Deployment

Docker (Standard)

docker build -t safeseedops -f docker/Dockerfile .
docker run -p 8000:8000 --env-file .env safeseedops

Or with Docker Compose:

docker-compose up

Docker (AMD ROCm)

For local inference on AMD GPUs with ROCm:

docker build -t safeseedops:rocm -f docker/Dockerfile.rocm .
docker run -p 8000:8000 --env-file .env --device=/dev/kfd --device=/dev/dri safeseedops:rocm

Render (Recommended)

Deploy on Render's free tier — auto-deploys from GitHub on every push.

1. Go to dashboard.render.com → New Web Service
2. Connect your GitHub repository, branch hf-gradio-sdk
3. Set:
   • Build Command: pip install -r requirements.txt
   • Start Command: python app.py
4. Add environment variables (see table below)
5. Deploy — the app is live in ~3 minutes at https://<name>.onrender.com

Environment Variables

Variable	Required	Default	Description
GOOGLE_API_KEY	Yes*	—	Google Gemini API key (primary)
FIREWORKS_API_KEY	No	—	Fireworks AI API key
GEMINI_API_KEY	No	—	Alias for GOOGLE_API_KEY
OPENAI_API_KEY	No	—	OpenAI API key (fallback)
ANTHROPIC_API_KEY	No	—	Anthropic API key (fallback)
DATABASE_URL	No	sqlite:///data/safeseedops.db	Database connection string
REDIS_URL	No	—	Redis URL for caching
LOG_LEVEL	No	INFO	Logging level

* At least one API key required.

---

Documentation Home

For complete guides, tutorials, design papers, and API specifications, visit the Documentation Home.

• Developer Setup Runbook
• Overall Architecture Index
• SafeSeedOps Pro Roadmap & Deferred Backlogs
• General Availability Release Notes

---

License

This project is licensed under the Apache License 2.0. See LICENSE for details.