2026 – now
Captain
Freelance Head of Engineering · AI-native FMCG analytics SaaS
- Lead a five-person engineering and data team. Introduced design docs with decision records for architecture, and spec-driven delivery with review depth scaled to risk and an independent second-model review on every PR. Shipped ~390 commits in five months.
- Designed and built the multi-agent analysis engine (Claude agents over Databricks) end to end. Interviewed users, defined the eval set and wrote the architecture: truth rules across state transitions, message transport, context rotation and compaction, and cancellation semantics. Designed the UI and shipped the product.
- Designed and built the agent-generated, human-clickable dashboards product on top of the multi-tenant SQL warehouse. Set trust boundaries between gateway, dashboard and query layer: isolated origin and strict CSP for dashboards, a session-checked gateway for every frontend filter callback, queries pinned to immutable versions with bound parameters, and invite-only access through revocable magic links.
- Replatformed production onto declarative infrastructure using Terraform, with keyless CI and a versioned, trunk-based release train. Retired legacy VMs and self-hosted CI.
- Put 15+ external systems behind one typed client layer, so credentials, retries and failure handling are defined once for backend and pipelines.