I help companies put a governed layer between their people and the AI models they use — one gateway for access, budgets, guardrails and tracing — assembled from proven open-source parts rather than built from scratch.
A single entry point for every model your company uses. Identity-based keys, per-team budgets and rate limits, screening for PII and leaked secrets, and a complete trace of every request. Access is granted deliberately, never assumed.
Tool servers, per-run sandboxes, and short-lived scoped credentials. Each agent gets exactly the tools, budget and lifetime its policy allows — and the policy lives in the repository, reviewed like any other change.
Kubernetes, Terraform, workload identity, managed secrets, internal services behind SSO. Infrastructure described in code, applied through pull requests, and reproducible from an empty cloud project.
A short, fixed-scope engagement: what you are running today, what it costs, where the exposure is, and a prioritised plan you can execute with or without me. Useful before committing to a larger build.
Most of what a platform needs already exists as mature open source. The work is choosing well, wiring it correctly, and keeping the glue small enough that your team can own it after I leave.
Every behaviour the platform depends on is backed by a check that runs. Vendor documentation is a starting point for investigation, not evidence — often enough, it turns out to be wrong.
Nothing important lives only on someone’s laptop. The environment can be torn down and rebuilt from the repository, which is what makes a handover real rather than ceremonial.
Budgets, rate limits and per-team spend attribution are part of the design, not something bolted on after the first surprising invoice.
A governed AI capability layer for the whole company, routed through a single gateway: access to both commercial and open models, identity-based keys with per-team budgets and rate limits, guardrails for personal data and credential leakage, and end-to-end tracing of spend and usage.
Alongside it, an internal chat portal behind single sign-on, a set of tool servers with per-team grants, and a sandboxed runner for autonomous agents. Delivered on managed Kubernetes with Terraform and Helm, and handed over with written documentation and a test suite that verifies the behaviour the platform relies on.
Further engagement details available on request.
For new engagements, a short note on what you are running today and what you are trying to change is enough to start.