A production-shaped customer-service agent: it reads intent, calls the right tools, queries live order data, retrieves policy from a RAG knowledge base, and answers — all behind enterprise guardrails.
Five stages take the project from empty repo to a working conversational agent. Click any stage to open its playbook — how it works, the stack layer, and the resources behind it.
The whole agent lives in one well-organised repo. Expand a folder to explore it, or click any file to see what it does — straight from the project structure.
The agent interprets each message, decides whether it needs structured data or unstructured knowledge, and routes itself to the right tool.
“Where is my order ORD-1001?” → the agent calls the Order tool, which runs a validated SQL query against PostgreSQL and returns live status.
“What is the cancellation policy?” → hybrid search over Qdrant retrieves the right policy chunk and the agent answers, grounded in source documents.
“Can I cancel order ORD-1001?” → the agent combines order state and policy, then routes sensitive actions through human-in-the-loop review.
Every layer maps to a well-understood, deployable technology — no bespoke magic.
For every decision point we compared the leading options on pros, cons and ideal use — then committed to a recommendation. Switch tabs to explore each layer; the recommended choice is starred and justified below each set.
Pick an utterance and run it. The request animates through each layer — UI → API → orchestrator → tool → data — with the trace in the console.
A phased path from the working prototype to a governed, multi-domain assistant.
Technical architecture pitch · built from the project structure deck.
Subtitle
Try a question