Complex systems. Measurable outcomes.

Challenge

Adjusters spent most of their first day on each claim gathering documents, checking coverage and chasing missing information across email and three core systems.

What we did

We built a supervised multi-agent system with MCP connectors to the policy, claims and document systems. Agents classify the claim, verify coverage, request missing items and prepare an adjuster brief. A person approves every customer-facing action.

Challenge

A promising copilot prototype had stalled for two quarters: no evaluation process, unclear cost per user and concerns about customer data isolation.

What we did

A Vidura forward deployed pod joined the product team, introduced tenant-isolated retrieval, an evaluation harness in CI and model routing to control cost, then handed the system over to the in-house team.

Challenge

Noisy alerts, unclear service ownership and no visibility into the latency and cost of newly launched AI features.

What we did

A unified OpenTelemetry observability stack, SLOs per user journey, AI-specific dashboards for token cost and quality, and an incident practice with clear ownership.