ArNa Resolve: Live AI Pipeline Demo

Type a support query into the chat. It will be classified by intent, checked for PII, matched against the knowledge base via RAG, and responded to by GPT-4o-mini, all in real time on Azure.

ArNa Resolve shows how I think about applied AI in production, from orchestration, retrieval, and escalation through to observability and governance. It is more than a chatbot front end.

Live

What happens when you send a message

1
PII Interception: your message passes through an Azure Function proxy that redacts any personal data before it reaches the LLM
2
Intent Classification: GPT-4o-mini classifies the query: billing, connectivity, device, general, or escalation-required
3
RAG Retrieval: Azure AI Search retrieves the most relevant knowledge base chunks using vector embeddings
4
Response Generation: a grounded answer is generated using the retrieved context. If confidence is low, the query escalates to a human agent via Slack
5
Audit Logging: every interaction (including PII flags) is written to Cosmos DB and visible in the governance dashboard

Things to try

Ask about connectivity issues, billing questions, or device setup. Try including a fake email or phone number to see PII interception in action.

e.g. "My router keeps dropping, I've rebooted it twice"  ·  "Can you help me with my bill, my email is test@example.com"  ·  "How do I set up a new device?"

Explore the stack

Tech stack

Azure Logic Apps Azure OpenAI GPT-4o-mini Azure AI Search Azure Functions Cosmos DB APIM App Insights Slack HITL
AR
ArNa Resolve
Agentic support AI
Hi, I'm the ArNa Resolve support AI. Ask me about connectivity, billing, device setup, or anything else. I'll do my best to help.
Messages are processed via Azure OpenAI · PII is intercepted and redacted