Testbusters

Problem

Testbusters is an eCommerce learning brand in the medical-education space: it prepares students for the exams that gate their careers, and sells the courses, books and tools that go with that. When I took over the digital product side, the store ran on legacy WooCommerce. It worked, but it leaked at the funnel and was hard to scale, and most of the operation behind it (post-purchase steps, catalog feeds, internal questions) still ran on manual work and constant cross-team coordination.

The migration

The spine of the work was a headless replatforming. I led the end-to-end move from WooCommerce to a composable stack: BigCommerce for commerce, Next.js for the storefront, Builder.io for content, covering catalog, cart and checkout, with zero downtime through the cutover.

Headless was not the goal in itself. It was what let me fix the funnel where people actually dropped (payment options, bundling and pricing, the steps that mattered), reach mobile-first performance, and give the marketing team autonomy to ship without waiting on engineering. I put GA4, a clean data layer and Looker Studio in place so each decision had a number behind it rather than an opinion.

Beyond the storefront

Post-purchase automation

Post-purchase operations ran on a chain of manual steps. I built one automation layer to replace them, as a modular n8n architecture: reusable components, shared helpers and standardized payloads, so each new flow builds on what already works. Reliability was designed in from the start: idempotency and deduplication, retries with backoff, and monitoring, logging and alerting. It sits over the APIs and webhooks as an integration layer, with data normalization to keep everything downstream consistent.

Internal AI assistant

Repetitive internal questions were eating real PM and ops time. I built a RAG pipeline end to end: chunking strategy, metadata, embeddings and vector search, with prompt and context management and evaluation loops on quality and relevance. Guardrails keep answers grounded in the knowledge base and cut hallucinations, so the team could answer its own questions instead of waiting on a person.

Multi-brand product feeds

Multi-brand catalogs needed consistent feed quality and a way to onboard new brands without starting from scratch each time. I built a standardized normalization layer (taxonomy rules, validation checks, brand-specific mapping) feeding Google Merchant Center and AWIN, so onboarding a new brand became a configuration step rather than a rebuild.

Outcome

The migration lifted conversion 27% year over year, cut cart abandonment 34%, and raised average order value 15.1%, with better scalability and mobile-first performance behind it. The post-purchase automation went live in two weeks against a month planned, and the RAG assistant gave back roughly three hours a day that used to disappear into repetitive support.

The work I am proudest of from this period was not on the storefront at all. It was Staff Manager, the internal platform I built to run the people side of the operation. That one has its own page.