Principal engineering through fast AI product discovery
I joined ChatSheet AI as Principal Engineer during a year of focused product experimentation. The team was exploring where AI could create practical leverage for users: not as a thin chat wrapper, but as backend-supported workflows that could search, retrieve, transform, and act on real business context.
That meant the engineering work had to support speed and change. Product assumptions were still moving, but the systems underneath could not be treated as disposable. Each MVP needed enough structure to prove the idea, support real usage, and give the team something reusable for the next iteration.
Making context searchable across CMSs
A major part of the work was enabling context search through vector databases across multiple CMS platforms. The product needed to understand and retrieve relevant content from different sources, then make that context available to AI workflows in a useful way.
I worked on the backend patterns behind that flow: ingesting content, representing it for retrieval, connecting it to vector search, and exposing it through product surfaces that could support LLM-driven workflows. The goal was to make context feel accessible to the product without forcing users or the team to manage the complexity of the retrieval layer directly.
Shipping multiple MVPs without losing the system
Across the engagement, I helped ship multiple MVPs while keeping the backend adaptable. Some work leaned into Go, some into Ruby on Rails, and some into the integration layer between AI services, CMS data, and product workflows.
My role was to stay close to both architecture and implementation. I shaped service boundaries, data flows, and integration decisions, then wrote the code needed to get those decisions into production. That balance mattered because the team needed momentum, not architecture documents that sat apart from the build.
The engineering tradeoff
The core tradeoff was speed versus continuity. In early-stage AI products, it is easy to move quickly by building each experiment as a one-off. That works until the team finds traction and has to carry the system forward.
For ChatSheet AI, I focused on backend choices that preserved learning. The systems supported fast MVP delivery, but they also gave the team reusable patterns for retrieval, integrations, and AI workflow infrastructure as the product direction evolved.