Embedded AI for a vertical SaaS platform
A mid-stage vertical SaaS platform serving operations teams in a service industry wanted AI capability inside their product on a defined timeline. They had no internal ML team and no appetite to build one before they knew which capabilities would matter to retention.
Fulcrum took the build end-to-end: a Python agent layer, a suite of MCP servers exposing the partner's product data model and a small set of external systems, a retrieval layer indexing customer-tenanted knowledge, and an eval harness scoped to the partner's customer-facing accuracy threshold. Production hardening included tenant isolation, structured tracing, and a feature-flagged rollout the partner's customer-success team could pace.
We carried the architecture decisions, the build, and the eval discipline. The partner kept the roadmap, the customer relationship, and the product surface — which is the shape that works when the partner is the company on the contract.