About
Mark Gerrard
I build and run the AI systems inside a UK telesales business — the multi-model orchestration layer, the agents that enrich and qualify leads, the verification and observability that keep them honest, and the internal tooling that ties it to telephony, CRM, and our planning-domain retrieval.
Everything I write about here is something I operate in production. I’m more interested in the engineering discipline around models — reliability, integration, failure modes — than in commentary about where the field is going.
Now Hardening the verification queue behind ~40k enrichment calls a month.
Work on
Agent reliability & verification
Multi-model orchestration
CRM enrichment pipelines
Planning-domain retrieval
Talks & projects
Reliability is a budget Talk · LLMs in Production · Feb 2026
Verification before autonomy Talk · PyData London · 2025
gate — a deterministic proposal verifier Project · internal, notes public
The review queue is the product Field note