🚀 Democratizing Synthetic Data for AI Builders

SynthForge is a privacy-first, lightweight tool that generates high-quality synthetic tabular data to overcome common AI bottlenecks like data scarcity, annotation challenges, and privacy concerns. Built as a Minimum Viable Product (MVP) in just 3 days, it's designed for indie developers, students, and teams with real-world constraints—making "infinite clean data" accessible without heavy compute or budgets.

Live Demo: synthforge.streamlit.app

In AI development, 80% of time is often wasted on data prep (Gartner). SynthForge flips that by enabling quick, ethical data generation:

• Solve Data Shortages: Create statistically similar datasets from small uploads.
• Prioritize Privacy: Built-in PII detection and differential privacy.
• Streamline Annotation: Auto-label with sentiment, binning, or clustering.
• Vision: Empower underrepresented builders (e.g., from Jaipur, India) to innovate globally.

Inspired by trends like synthetic data (market: $1-2B by 2025) and tools like Faker/SDV, but focused on simplicity and accessibility.

• Upload & Generate: Supports CSV/Excel (auto-samples large files for efficiency).
• Customization: Adjust variance for randomness, add differential privacy noise.
• Privacy Tools: Scans for emails/phones; optional epsilon-based anonymization.
• Auto-Labeling: Rule-based or LLM-enhanced (via free Groq tier) for sentiment on text, binning/clustering on numerics.
• Outputs: Download synthetic CSV + HTML report with stats comparisons (means, stds, top values).
• Lightweight: Runs smoothly on modest hardware—no GPUs required.

1. Clone the repo:

   git clone https://github.com/yourusername/synthforge.git
   cd synthforge
   

2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

3. Install dependencies:

   pip install streamlit pandas numpy faker scikit-learn openpyxl requests
   

4. (Optional) For LLM-enhanced labeling: Sign up for a free Groq API key at groq.com and input it in the app sidebar.

Run the app locally:

streamlit run app.py

• Open in your browser (defaults to http://localhost:8501).
• Upload a file, tweak settings in the sidebar, and generate!
• For production: Deploy to Streamlit Sharing or similar (as done for the live demo).

Example: Upload a CSV with names, emails, ages—watch it generate synthetics while masking PII.

• Frontend: Streamlit (simple, interactive UI)
• Core Libraries: Pandas (data handling), NumPy (stats), Faker (heuristic generation), scikit-learn (clustering/imputation)
• Privacy/Labeling: Regex for PII, Laplace noise for DP, Requests for optional Groq LLM
• Deployment: Streamlit Cloud (free tier)

• v1.1: Multi-modal support (text/images).
• v1.2: API endpoints for integrations (e.g., Jupyter/HF).
• v2.0: Federated privacy, compute optimization, bias auditing.
• Long-term: Agentic workflows and a synthetic data marketplace.

We welcome contributions! See CONTRIBUTING.md (add if needed).

Fork the repo, create a branch, and submit a PR. Focus areas: Bug fixes, new heuristics, modality expansions. Let's build together!

MIT License – See LICENSE for details.

• Hanish (Founder): LinkedIn | [email protected]
• Issues/PRs: GitHub Issues
• Feedback: Test the app and drop thoughts on LinkedIn or X!

Built with ❤️ from Jaipur, India. Join the forge—let's crush AI bottlenecks! 🚀