This project is an Enterprise Autonomous AI Orchestration & Evaluation Pipeline designed for customer support.

A concise overview of its architecture and capabilities:

Core Capabilities
Hybrid Routing Engine: Uses a fast-track heuristic router to bypass the LLM for simple queries (resulting in <0.1s response latency) and redirects complex requests to a Google Gemini-powered Unified Agent with native function-calling capabilities.
Privacy & Compliance (GDPR): Features a PrivacyScrubber interceptor layer that reversibly pseudonymizes PII (names, addresses, phone numbers) before data is sent to external LLM endpoints.
Automated Evaluation Suite: Includes a python-native benchmarking suite to run regression testing against adversarial inputs to guarantee routing accuracy, safety, and latency standards.
Real-time Observability: Emits detailed JSON telemetry metrics mapping token usage, agent execution paths, and tool latency.
B2B SaaS Multi-Tenancy: Designed with PostgreSQL Row-Level Security (RLS) and namespaced vector databases to support data isolation between multiple client organizations.
Tech Stack
Backend: FastAPI (Python), Google AI SDK (Gemini Flash), FAISS Vector Index (RAG).
Frontend/Admin: Next.js (TypeScript, Tailwind/Shadcn, Framer Motion) featuring an e-commerce customer interface and an administrative analytics dashboard.
Data & Infrastructure: Supabase PostgreSQL, Docker / Docker-Compose, and AWS EC2.