Fractional CTO: what it is, what it costs, and when to hire one
What a fractional CTO does and costs, the risks, and how to tell whether you need a CTO, a Chief AI Officer, or neither.
Read →Fractional Chief AI Officer: what it is, what it costs, and when to hire one
What the role owns, the first thirty days, how you will know it works, what it costs, and the four cases where hiring one is a mistake.
Read →What a forward deployed engineer does inside your company
Written for business owners: the first month, the decisions that are yours, what you will never be asked to do, and what your team owns at the end.
Read →AI in retail: what gets built inside a retail chain
Nineteen months building a retail chain's operating platform. The reconciliation before any model, the forecast done by hand, the order in which to try models, open-to-buy, and where AI is not worth it.
Read →Forward deployed engineer vs solutions engineer vs sales engineer vs solutions architect
Four roles that get confused with each other. What each one owns, when each one works, and which one you need.
Read →Forward deployed engineering: how an engagement works, week by week
Week by week, from the records: the decisions file, the weekly shipping rule, the compiler checks, the mock services, the platform limits, the handover order, and when the model does not fit.
Read →What is a forward deployed engineer?
The first week, the mistake a local reviewer catches that no engineer would, the handover, and when hiring one is a mistake. From someone who does the work.
Read →Tiny Model, Big Jobs: Optimizing LLMs with DSPy and GEPA
An 8B model on a warehouse-robot task goes from 30 percent to 95 percent with a prompt GEPA wrote from its own failures, beating a model thirty times its size. The deck shows the recorded run; the notebook runs it again on your machine.
Open the deck →Self-hosted LLM: how to keep your company's data private
What data leaves the company in a normal deployment, the four ways to keep it inside, sizing by workload, what it costs to run, and when self-hosting is not worth it.
Read →Running Gemma-4 26B at 124 tokens/sec on a CPU, no GPU
A 26-billion-parameter mixture-of-experts model on a plain desktop with no graphics card. The recipe, the byte budget, and where the real speed limits are.
Read →