Ecommerce at real traffic: 40+ NordicTrack and ProForm sites, in-house services talking daily to vendor stacks.

• Led monolith → microservices so teams could ship and scale without stepping on each other.
• Owned a mostly serverless AWS backend for orders, SKUs, catalog data, and the logs you actually read when things break.
• Wired services to Commerce Layer, Contentstack, Algolia, and SAP with contracts people could trust.

Regulated customer service: AI self-service for members, tools for agents, audits breathing down your neck.

• Chatbots with NLP, LUIS, and Microsoft Bot Framework, with guardrails included.
• Drag legacy UIs forward with TypeScript/React talking to REST services instead of hope.
• Shipped on Azure and AWS with HA patterns; Spring/J2EE still in the mix where it belonged.

Microservices that had to stay up, plus multi-cloud deployment when one vendor wasn’t enough.

• Spring Boot services, Cassandra/Hibernate underneath, React SPAs on top.
• Kubernetes on AWS and Docker-based pipelines so deploys weren’t artisanal.

• API calls are authenticated at the edge; Lambdas stay small and typed validation is shared so the browser and server don’t argue in production.
• Résumé uploads use presigned URLs and an archive bucket so the old file isn’t lost when a new one goes live. Keys never sit in the client.
• CORS, WAF-style hardening in prod, secrets in a vault, CI that uses short-lived cloud login: same habits as a real shipping system, just smaller scope.
• Light analytics and contact capture with sane rate behavior; traffic is modest today, but the model doesn’t paint me into a corner later.

• RBAC for admin, instructor, and learner so permission checks stay in one place as more orgs come on board.
• Serverless on AWS: scales on demand, tight IAM, partition keys chosen so enrollment spikes don’t melt one shard.
• APIs are contract-first for web and mobile: versioning, paging, idempotent writes so bad Wi‑Fi and bulk imports don’t corrupt state.
• Per-tenant config and a sane path for schema change so new customers don’t mean a forked codebase per school.

• Prototyping real-time flows against cold starts, connection fan-in, and how far managed AWS can stretch before you reach for something heavier.
• Honest notes on modular monolith vs event slices, with tests and perf budgets before a feature laundry list.
• Details stay private for now; the focus is scale, safety, and how you’d operate it day one.

• Split services so chatty real-time paths don’t drag down stable catalog/booking work; each slice gets a failure budget.
• Mobile-first APIs: idempotent writes, friendly backoff, and indexing choices held until we stress-test peak season.
• Public write-up stays thin; internally it’s about observability, rollout gates, and staged traffic (NDA until ship).

• Listings and categories modeled so search stays quick as inventory grows, without N+1 headaches on skinny mobile responses.
• Filters and pagination behave the same no matter which client hits the API; indexing notes are ready if we shard or add replicas later.
• Trust signals and abuse-resistant posting were part of v1, not something we bolted on after the first incident.

• Split entitlements from catalog so pay-per-view vs subscription logic can change without breaking every client.
• Media access through short-lived signed URLs and session patterns that can handle abuse; storage/CDN sit behind stable APIs.
• Metadata and search tuned for burst traffic, with caching and query shapes so premiere night doesn’t flatten the database.
• Latency and rights called out in design docs; licensing minutiae stay off the public site.