ExamWorks: parachuting into an in-flight program — 80+ build-ready stories in one week.

ExamWorks is a national provider of independent medical examinations — a business that runs on documents. Referral letters, exam reports, notices, invoices: a library that had grown across multiple acquired brands into 80+ overlapping source documents, each a variation on the others, none governed.

I joined the Salesforce program mid-flight, in a solution-consultant role — developers already sprinting, documentation sparse, and a backlog that wasn't keeping up with the build team's appetite. There was no runway for a six-week discovery. The program needed someone producing build-ready work in days.

• Feed the build team. Developers were outpacing the backlog. Every sprint without ready stories was paid-for capacity doing nothing.
• Tame the document library. 80+ fragmented, multi-brand templates needed to become a scalable document-generation system — with an architecture and governance model, not just fewer files.

The traditional consulting answer — weeks of stakeholder interviews before producing anything — wasn't available. The question became:

I built the ramp-up itself as a system: every scrap of program documentation went into an AI-readable knowledge base, then Claude Code drafted against it while I supplied the judgment — what's right, what's missing, what a developer actually needs to see in acceptance criteria.

• Knowledge base first — program docs, org metadata, and meeting notes consolidated into NotebookLM so every AI draft was grounded in the client's reality, not generic patterns
• Claude Code as the drafting engine — user stories generated from sparse documentation across Billing, Revenue, and document generation, complete with acceptance criteria in the program's own format
• Judgment stays human — every story reviewed, corrected, and stress-tested against what I knew of the org before it touched the backlog
• Documentation became infrastructure — I set the standard for the program's Confluence page setup, so the knowledge that powered my ramp-up became a durable, structured asset for the whole team
• The method scaled beyond me — I authored the install and getting-started guidance and set up colleagues on Claude Code, making AI-assisted delivery a team capability instead of a personal trick

By the end of the first week, the backlog had 80+ accurate, build-ready user stories spanning Billing, Revenue, and document generation — three domains, ramped cold, with enough runway that developers never idled waiting on requirements again. What normally takes a consultant a month of ramp-up happened inside a sprint.

In parallel, I owned the document-generation strategy. The answer to 80+ overlapping templates wasn't deleting files — it was an architecture that made variation cheap and sprawl impossible:

• Tokens for every piece of variable data, so one template serves many contexts
• Conditional blocks for brand- and state-specific language, replacing whole duplicate documents
• Custom metadata driving assembly, so business users configure instead of clone
• Exception governance — a process for when someone wants template #15, so consolidation survives contact with the future

I ran the design workshop that aligned stakeholders on the model and authored the 15-story delivery set that took it into build.

Beyond the engagement, this became my calling card internally: the go-to example of what AI-assisted delivery looks like when it works — not AI replacing the consultant, but AI collapsing the distance between joining a program and contributing to it.

Ramp-up is a design problem. Nobody designs their own onboarding — they just endure it. Treating "get productive fast" as a system to engineer, with a knowledge base and a drafting pipeline, turned a weakness of consulting (the cold start) into an advantage.

AI moves the bottleneck to judgment. Drafting 80 stories was never the hard part once the machine did it — knowing which 80, catching what the docs didn't say, and writing acceptance criteria a developer trusts: that stayed human, and that's where the value concentrated.

Governance is what makes consolidation stick. Anyone can merge 80 documents into 14 once. The exception process — what happens when someone wants #15 — is the difference between a cleanup and an architecture.