Show HN: Misata – synthetic data engine using LLM and Vectorized NumPy

Summary

🧠 Misata Generate realistic multi-table datasets from natural language. No schema writing. No training data. Just describe what you need. ✨ What Makes Misata Different Feature Faker SDV Misata Natural language input ❌ ❌ ✅ Auto schema generation ❌ ❌ ✅ Relational integrity ❌ ✅ ✅ Business constraints ❌ ❌ ✅ No training data needed ✅ ❌ ✅ Streaming (10M+ rows) ❌ ❌ ✅ 🚀 Quick Start pip install misata With Groq (Free, Fast) export GROQ_API_KEY=your_key # Get free: https://console.groq.com misata generate --story " A SaaS with 50K users, subscriptions, and payments " --use-llm With OpenAI export OPENAI_API_KEY=your_key misata generate --story " E-commerce with products and orders " --use-llm --provider openai With Ollama (Local, Free, Private) ollama run llama3 # Start Ollama first misata generate --story " Fitness app with workouts " --use-llm --provider ollama 📊 Example Output $ misata generate --story "A fitness app with 50K users" --use-llm 🧠 Using Groq (llama-3.3-70b-versatile) for intelligent parsing... ✅ LLM schema generated successfully! 📋 Schema: FitnessApp Tables: 5 Relationships: 4 🔧 Generating 5 table(s)... ✓ exercises (10 rows) ✓ plans (5 rows) ✓ users (50,000 rows) ✓ subscriptions (45,000 rows) ✓ workouts (500,000 rows) ⏱️ Generation time: 2.34 seconds 🚀 Performance: 213,675 rows/second 💾 Data saved to: ./generated_data 💻 Python API from misata import DataSimulator , SchemaConfig from misata . llm_parser import LLMSchemaGenerator # Generate schema from story llm = LLMSchemaGenerator ( provider = "groq" ) # or "openai", "ollama" config = llm . generate_from_story ( "A mobile fitness app with 50K users, workout tracking, " "premium subscriptions, and January signup spikes" ) # Generate data for table_name , batch in DataSimulator ( config ). generate_all (): print ( f"Generated { len ( batch ) } rows for { table_name } " ) 🔧 CLI Reference # Basic generation (rule-based, no API key needed) misata generate --story " SaaS company with users and subscriptions " # LLM-...