CREAMLABS AX

At CREAMLABS, I independently designed, delivered, and operated business systems and LLM-enabled workflows. The work spans customer-facing commerce and content platforms as well as internal operations. Each engagement started with the existing process: how people maintained information, where they searched for it, and which decisions still required a person.

The public interfaces bring product discovery and purchase options together with editorial content, community records, and artist portfolios. These different surfaces required attention to the relationship between structured information and visual material. A product option, a published post, and a portfolio entry each need a clear path from browsing to the next action.

For internal workflows, I separated the LLM service into a FastAPI layer alongside Spring Boot applications. Function calling connected the model to internal APIs, and JSON Schema validation controlled what could enter a business workflow. Each business indexed roughly 10,000 internal documents for retrieval, with about 2,000 to 3,000 search requests per day on average.

Operating the systems meant balancing model cost and latency while keeping outputs inspectable. I used model routing and regression evaluations, then monitored response quality through schema checks and execution logs. Rule-based safeguards rejected unsuitable results. Failed outputs could be regenerated or sent for administrator review, keeping a human decision available when the workflow needed one.