Smart. Language. Programming.
Services
Fractional AI/ML Lead
How it works
Two days per week, embedded with your team.
The work defines the roles before you commit to the size of the team. You get someone with 20 years of putting systems into production and keeping them there.
The gap
You need senior AI/ML judgment but can't fill the role, or you can't justify a full-time hire yet.
AI work is already happening, driven by those who are enthusiastic, not by someone who knows the pitfalls of a deployed system.
The Terms
Renewable four-month terms, priced within the budget for the hire.
No time lost getting a new employee up to speed, so you see real progress within weeks.
AI/ML Readiness assessment
The problem
You can usually spot it from three things.
The demo works and production doesn't.
Nobody can tell you which systems are load-bearing and which are just sitting there.
The roadmap is a list of technologies instead of a list of outcomes.
How I Think About it
LLMs are extremely valuable tools. They're just not what everything calls for, and the gap between those two things is where most of the money gets wasted.
The right model in the right place. Where digging and reasoning need to happen, an LLM shines. Where the answer is computable, it doesn't.
What you get
A hands-on review. Not a stakeholder survey. I look at the data, the models, the pipelines, and the tooling, including the parts that never come up in meetings.
An honest capability read.
A twelve-month roadmap.
A readout for two audiences. Your engineers need the technical reasoning. The people paying for the work need the shape of the decision.
Four weeks
Fixed Fee $30K
About
Stephanie Poisson builds AI systems on a foundation many in the field skipped: the science of language itself, and not just English. A B.A. in English Literature led to an M.S. in Computational Linguistics at Georgetown, and to a career-long conviction that genuinely understanding language, its structure, its ambiguity, and its stubborn refusal to behave, is the foundation on which useful language technology gets built. She has spent twenty years proving that conviction in production, well before “AI” became everyone’s favorite buzzword.
The proof has a paper trail. The first system she shipped, a coreference and relation-extraction prototype for English and Arabic, built as a summer intern in 2005, was released to analysts for operational use. Over the following decade at the National Security Agency, she led a multilingual event processing and reasoning project from early prototype to a full-scale, web-based analytic used by customers, directed annotation efforts across five languages, published in lexical semantics, and built NLP systems spanning nine-plus languages, designed algorithms extensible to new languages through a parameter file alone. She also led people, not just projects. As Associate Director and Hiring Manager for Scientific Linguists, she rebuilt the Agency’s development program for technical linguists, created its first interview guide, and taught the courses her successors learned from.
In industry, the pattern held. The four recommender systems she built between 2019 and 2024 to make U.S. government data searchable and interoperable are still in production today, a quiet signature of work done right.
That arc, classical statistical and linguistic methods through the modern LLM stack, research to production, gives her clients a rare thing: a consultant who matches the technique to the problem rather than the trend, and who has receipts going back two decades. Simply put: what she builds ships, and what she ships stays running.