Confidentiality: this page describes the approach and the
categories of work only. No customer data, internal designs or proprietary business rules
are included.
The problem
An engineering team shipping a rules-heavy enterprise platform was spending a large share of
every sprint on work that was necessary but repetitive: writing RFCs for each significant
change, keeping functional test scenarios in step with evolving business rules, and producing
technical documentation that lagged behind the code.
Rebate logic is intricate — eligibility, thresholds, tiers, periods, partner-specific
agreements — so gaps in tests and documentation translated directly into regression risk and
slow onboarding.
Constraints
- A production enterprise system with real financial exposure. Nothing
generated could be trusted without review.
- The AI had to work inside the team's existing tools and review flow,
not replace them.
- Consistent output format mattered more than clever output. Reviewers
needed to know where to look.
What I designed and built
- RFC generation skill. A Claude Agent Skill that takes a requirement or
ticket plus pointers to the relevant services and produces a structured RFC: context,
problem statement, options considered, proposed design, data and API impact, rollout plan
and open questions. Engineers review and edit rather than start from a blank page.
- Functional test scenario generation. A skill that reads a requirement
and the existing scenario set, then proposes new and updated scenarios in the team's
format — flagging which existing scenarios the change is likely to affect.
- Engineering assistants for daily work. Requirements analysis, code
review assistance, troubleshooting and technical documentation, all constrained to the
team's conventions.
- Platform architecture. In parallel I owned the microservices and
event-driven design (Kafka for asynchronous business events), API security, observability,
CI/CD, and the SQL and NoSQL data models the AI tooling had to understand.
Stack
Claude on AWS · Claude Agent Skills · microservices · Kafka · REST ·
PostgreSQL and NoSQL stores · Docker · Kubernetes · CI/CD · React
Outcomes
- RFCs and test scenarios produced from a consistent template in minutes, then reviewed,
instead of written from scratch.
- Regression coverage improved, because scenario maintenance became a review task rather
than an authoring task.
- Repetitive documentation work moved off the critical path of every sprint.
- The same skill pattern was reused across services — which is what makes it a product
rather than a one-off.
What this means for your team
This is the engagement I now sell as Engineering AI Enablement: Claude
Code, Agent Skills and MCP integrations set up around your actual repositories, tickets and
conventions, so your engineers get reviewable drafts of RFCs, tests and docs instead of
blank pages. About four weeks to set up, with ongoing monthly support if you want it.