Case study

Merative

The engagement combined platform migration with a carefully bounded AI workflow: improve how traffic reaches the web estate, then improve how a business-critical form is reviewed.

Client

Merative

Role

Edge migration and AI workflow engineering

Year

2024–2026

Summary

Helped migrate a large web estate from Cloudflare to Azure Front Door and create an AI-assisted workflow that evaluates every submission to an important inbound form.

The brief

My work with Merative covered two important but related areas: helping move a large web estate from Cloudflare to Azure Front Door, and helping create an AI agent that monitors a high-value inbound form.

Both workstreams required careful handling of boundaries. The migration had to preserve web behavior while changing a critical infrastructure layer. The agent had to add useful validation context without turning uncertain public information into an automatic business decision.

Migrating the edge layer

Moving from one edge platform to another is not a DNS-only change. Routing, certificates, caching, redirects, origins, security controls, observability, and rollback all participate in the cutover.

I helped approach the migration as a sequence of verifiable behaviors:

  • inventory the domains, routes, origins, and existing edge responsibilities;
  • translate behavior into Azure Front Door configuration;
  • validate certificates, redirects, caching, and origin health;
  • compare responses before traffic moved;
  • plan bounded cutovers and recovery paths; and
  • monitor the new edge after activation.

This reduced the risk of treating a broad estate as one undifferentiated switch.

The form-validation agent

The second workstream focused on an important form where every submission deserved review. The agent examines submitted context, searches public web sources for relevant information about the person or organization, and assembles evidence about whether the identity and stated reason for contacting Merative appear credible.

The system is best understood as evidence-assisted triage. It helps reviewers start with organized context rather than an isolated form payload. It does not make public-web information inherently authoritative, and this case study does not disclose prompts, thresholds, providers, personal data, or bypass conditions.

Designing the boundary around AI

The useful output is not a confident sentence. It is a traceable summary that makes uncertainty visible.

The workflow was shaped around:

  • processing every submission consistently;
  • separating submitted claims from externally discovered context;
  • preserving source and reasoning signals for review;
  • handling missing or contradictory evidence;
  • minimizing unnecessary exposure of personal information; and
  • keeping business judgment outside the model itself.

The takeaway

This work demonstrates two kinds of platform thinking. The Azure Front Door migration required disciplined infrastructure translation and cutover planning. The agent required equally disciplined decisions about evidence, uncertainty, and human ownership. In both cases, the strongest design came from making boundaries and recovery paths explicit.

Related writing