About Sigmoidal
We build AI that has to work
Sigmoidal is an AI consulting and engineering firm founded in 2016, years before the LLM wave made AI a board agenda item. We embed senior teams inside enterprises, life sciences organizations, and private equity firms, and we stay until the system proves itself in the operating numbers.
The short version
Sigmoidal opened in 2016 building machine learning systems for clients who needed them to run, not to demo well. The early engagements set the pattern that still defines the firm: a small senior team, embedded with the client, accountable for a production system and the number it moves.
The technology underneath has changed several times since: computer vision, classical ML, NLP, and now LLM-based systems. The standard has not. Over time the work concentrated in three domains, because clients in those fields kept asking us back: Enterprise AI Deployment, Life Sciences & Healthcare, and Private Equity & Portfolio Value Creation. The life sciences work includes Novartis, where we built predictive analytics for drug sales.
Method
Assess. Build. Operate.
Three phases, one accountable team, a number attached to each.
Assess
We start with the P&L, not the data lake. In two weeks, we identify where AI moves a real financial or operational number, size the effect, and kill the ideas that will not survive production. The output is a ranked plan with owners, dependencies, and an expected value on every line.
Build
An engagement lead, three to five senior engineers, and a domain specialist embed with your team and ship to production. We build to production standards from the first commit: versioned data, evaluation harnesses, audit trails, security review. A working system is in front of users by week six.
Operate
Shipping is the midpoint. We run the system, measure it against the number we committed to, and harden it against drift, edge cases, and scale. Then we transfer it: documentation, runbooks, and your engineers trained to own it.
Principles
How we decide
The system must run without us.
A demo is a hypothesis; a system carrying live decisions is a result.
Own the outcome.
We commit to the number, not to the effort.
Senior people only.
The engineers in the first meeting are the engineers who write the code.
Measured in client numbers.
Success is your metric, in your reporting, verified by your finance team.
Say no early.
If we cannot see the number moving, we say so before the engagement scales.
The team
The bench
Sigmoidal stays deliberately small: around forty people, most of them engineers, all of them expected to understand the number a system has to move, not just the code that moves it. The team includes alumni of Jane Street, Mozilla, Intel, and Capgemini; PhDs in machine learning and computational biology where the problem calls for one; and engineers who have shipped under GxP and financial-services scrutiny. Median engineering tenure is over seven years. There is no pyramid of juniors behind a partner's business card.
~40
People, most of them engineers
7+
Years median engineering tenure
Security
Trust, in writing
Client data stays in client-controlled environments by default, and where models must touch sensitive data, we work inside your cloud tenancy under your access controls. Penetration tests and vendor-risk reviews are routine, and our security lead answers your questionnaire personally.
- Client-controlled environments
- Routine penetration tests
Get a straight read
Bring the problem and the number it touches. We will tell you whether AI moves it, and what we would commit to if it does.