SchemaIntern -> DB-Aware NL to SQL

SchemaIntern is an enterprise-grade platform that bridges the gap between natural language and structured data querying. Unlike generic LLM wrappers, SchemaIntern is schema-aware — it connects directly to your database, discovers table structures, columns, relationships, and sample data, and uses that context to generate accurate, optimized SQL. The platform features a 10-layer guardrail system that prevents SQL injection, enforces read-only access, validates schema adherence, and estimates query cost before execution. Every query passes through classification, planning, generation, and policy enforcement stages before execution. Built with FastAPI (Python) + Next.js 15 (TypeScript) with a modular pipeline architecture, SchemaIntern supports PostgreSQL, MySQL, Snowflake, BigQuery, and DuckDB. The AI pipeline uses Qwen2.5-Coder-7B-Instruct via HuggingFace router (with vLLM/ROCm support on AMD MI300X for local inference). Key capabilities: multi-tenant with JWT auth, self-learning feedback loop, session management with Redis, vector storage with Qdrant, schema caching, and explainable AI responses. Frontend features real-time chat UI, schema browser, team management, and API key management.