Self-Correcting SQL Agent

An intelligent agent that writes SQL from natural language, executes it, and self-corrects using reflection + schema-aware prompting.

This demo showcases agentic AI principles:

• 🔄 Reflection Loop - Automatically retries with error context
• 🧠 Error-Aware Reasoning - Learns from database error messages
• 💡 Intelligent Correction - Not random retries, but thoughtful fixes
• 📊 Interactive UI - Streamlit dashboard showing the correction process

Perfect for demonstrating:

• LangGraph state machines
• LLM function-calling patterns
• Agentic workflows with error recovery
• Production-ready AI systems

✅ Agentic AI Capabilities

• State machines (LangGraph for workflow control)
• Reflection architecture (error analysis + correction loop)
• Tool use (database execution as a tool)
• Multi-step reasoning (generate → execute → reflect → correct)

✅ Production Thinking

• Thread-safe database handling (SQLite in Streamlit)
• Environment variable management (.env + secrets)
• Structured error handling (not just try/except)
• Observability (execution traces + metrics)

✅ LLM Engineering Skills

• Schema-aware prompting (dynamic context injection)
• Few-shot correction (error patterns in prompts)
• Rate limit awareness (max attempts configuration)
• Model agnostic (works with GPT-4, 3.5, or local models)

✅ Portfolio Value

• Complete working demo (not just a notebook)
• Interactive UI (proves concept visually)
• Self-contained (runs with streamlit run)
• Well-documented (recruiters can understand in 5 minutes)

This is the kind of project that gets you hired for Gen AI roles.