Parth106203/llm-scaffolding-architecture

🛡️ Deterministic AI Scaffolding Architecture

Production-grade AI Orchestration using Pydantic, Supabase, and ReAct.

This Hugging Face Space showcases a production-ready safety net around LLMs. Instead of granting database mutation access directly to the LLM agent, all operations output structured JSON intents which are checked by a deterministic Pydantic Validation Gate and self-corrected in a retry loop.

Core Features:

• —💬 Sandboxed Chat & RAG: Astronomy object catalog with ambiguity checks and verification.
• —📄 PDF Data Extraction: Real-time extraction of structured data from invoice PDFs with strict math validation.
• —⚖️ Comparative A/B Test: Parallel evaluation comparing scaffolded self-correction against raw, unscaffolded LLM parsing.