← All work
AI agentsActiveHealthcare AI

Custom GPT Suite

Ten-plus purpose-built assistants, each scoped to one real workflow.

A collection of 10+ production GPTs with engineered system prompts, curated knowledge bases and API actions. Each one is scoped to a single workflow rather than built as a general assistant, because a narrow agent with a defined failure mode is worth more than a broad one without.

At a glance

10+ production GPTs in active use
Engineered system prompts with enforced brand-voice and policy constraints
Curated knowledge bases plus API actions for outside reach
Explicit escalation logic for out-of-scope cases
Multi-language support built in

The approach

One job per agent. Each has an engineered system prompt, a curated knowledge base, and API actions where it needs to reach outside itself. Brand-voice and policy constraints are enforced in the prompt contract rather than hoped for, and every agent has explicit escalation logic for the cases it should not handle alone.

Why narrow beats broad

A general-purpose assistant has an undefined failure surface, which makes it impossible to evaluate and therefore impossible to trust with anything that matters. A narrow agent can be given a golden set, graded case by case, and improved against a measure. That is the only version of this work I will ship.

Built with

GPT-4Claude APIPrompt engineeringKnowledge retrievalAPI actions

Next case study

Workflow Automation Pipelines →

Cross-platform pipelines connecting CRMs, billing, email and comms.


Have something that should be automated?

Tell me the rough shape of the problem. I will tell you honestly whether it is worth building — and if it is not, I will say so.