Projects & engineering work.

Building AI workflows around the service problem

Start with the business problem

I use the CRM with a colleague for internal customer-service work. The design starts with practical questions: who needs a follow-up, what business context informs the next action, and who should approve it.

Shape the workflow and data

I built a shared data model, API and CLI around customer relationships, projects, invoices and versioned notes. Follow-up reminders are implemented. These provide the context and controlled write path for business-specific AI workflows.

Connect analysis to execution

The integration pattern is reminder or business event, contextual analysis, proposed action, human review, execution and a recorded result. Broader AI analysis and decision flows are extension areas; their rules and approval points must be tailored to the actual service process.

Make the workflow safe to operate

People retain approval of consequential actions. Version checks reject stale edits, retries do not create duplicate records, and note history preserves prior content. The next iteration should be driven by real usage and unresolved cases; no measured productivity gain or external adoption is claimed.