Use this when
The idea involves AI or complex software, and the real question is whether it can work reliably.
A workflow needs agents
Research, content review, archive exploration, creative production, or operations need tool-using AI with human checks.
The answers need sources
Retrieval, citations, timestamps, review states, and approved source material matter more than generic chatbot fluency.
The prototype may become real software
The architecture needs APIs, data pipelines, evaluation, monitoring, deployment, and a path to maintainability.
What ships
Enough engineering to make a real decision.
- Workflow map, user roles, source model, and success criteria.
- LLM agent, retrieval interface, browser-local AI demo, or review tool.
- Custom application, API, data pipeline, or integration layer.
- Evaluation cases, failure-mode review, and recommendation for production.
- Documentation, deployment notes, and handover for the team running it.
Related work
AI and software systems grounded in archives, public interfaces, and real operating constraints.
Send the workflow.
Email the users, source material, current process, approval constraints, and what a good answer or output must prove. The first step is usually a small prototype with real examples.
Email the workflow brief