FiveForty

01 — The problem

Support teams drown in repetitive tickets, and generic chatbots hallucinate — eroding the exact trust support is supposed to build.

02 — The decision

Ground every answer in retrieval, never the model's memory. A microservice mesh isolates the agent loop, RAG, and document storage so each can scale — and fail — independently.

03 — Architecture

• Microservice mesh: sync, ticketing, users, chat orchestration, RAG, and document storage
• RAG pipeline grounds every proposed resolution in the customer's own knowledge base
• Agent loop ingests tickets and proposes resolutions with humans kept in the loop
• Azure Container Apps provisioned with Pulumi, with separate dev and production environments
• Region chosen by service availability: workloads moved to France Central, where both the required Foundry models and Azure Redis were available
• Diagnosed BullMQ job buildup — broken TTL cleanup was driving container memory pressure

04 — Outcome

Tickets ingested, retrieved against, and resolved with cited, grounded answers — the anti-demo-ware AI platform.

Stack