No open roles, but read this.

We're a small operator-led firm. We hire when a customer engagement genuinely needs a hand ; not when a hiring plan says so. There are no open roles right now.

Who we'll talk to anyway

We keep an informal list of people we'd want to call when we do hire. If any of these describe you, send us a note ; even with no role open:

• LLM platform engineers who've shipped semantic caching, prompt caching, model routing, or batch-API integration in production.
• FinOps practitioners with experience reconciling cloud spend at 8–9 figures, ideally with an AI workload.
• ML quality engineers who've built A/B harnesses for generative output, including LLM-as-judge eval frameworks.
• Forward-deployed operators with consulting or solutions-architect chops at OpenAI, Anthropic, AWS, Databricks, Snowflake, or comparable.

How we work

• Remote-first, EU/US timezones. Async-default, weekly sync.
• Engagement-based delivery. You'll be on one customer at a time, not three.
• Profit-share, not equity-lottery. We don't have a venture round to make 0.05% feel meaningful.
• No sales quota. Technical work only.

Get in touch

Send a note + a single thing you've shipped: hello@llmcfo.com. No cover letter needed. No CV gymnastics.

What working here actually looks like

Engagements are short and specific. A typical one starts with a baseline period where the whole job is reconciling provider invoices against usage exports until the numbers agree, then moves into attribution, then into two or three optimizations that are measured against a later invoice. That first phase is unglamorous data work and it is where most of the value is created — every savings claim later in the engagement is only as credible as the baseline it is compared against.

The optimizations themselves are engineering, not advice. Prompt caching, batch routing, semantic caching, and model substitution all get implemented and A/B tested against a quality bar the customer sets before the test begins. Nobody here writes a recommendation deck and leaves.

What we don't want

• Generalists looking to learn AI on the job. Customers are paying for judgment that already exists; there is no bench to learn on.
• People who want to own a roadmap. The work is customer-shaped and changes every engagement.
• Anyone who needs a title. There is no ladder to climb here, and pretending otherwise would waste a year of your career.

What happens after you write

You get a reply either way, usually inside a week. If there is a fit, the conversation is about a system you have already built — how the numbers were reconciled, what broke, what you would do differently — rather than a take-home exercise. If there is no work to offer, we say so directly instead of keeping a note warm against a headcount plan that does not exist.

Contract and advisory arrangements are open even when permanent roles are not, where a defined piece of work exists: attribution instrumentation, invoice reconciliation, gateway migration, or building the eval harness an optimization needs.