Fractional data & AI lead
A few days a month of senior direction: architecture calls, review, and saying no to the expensive ideas early.
Consulting, staffing and data engineering for financial, insurance, retail and automotive companies — and AI services built for work where a person still has to sign the decision.
So we do both. The consulting practice puts the people and the data platform in place; the AI practice builds on top of it. We will tell you plainly when the second one should wait for the first.
People and systems. Staffing, remote and hybrid teams, data warehousing, business intelligence, and custom development for regulated and operationally heavy industries.
Claims and submission triage, document intelligence over policy wordings and filings, and analyst interfaces with an agent inside. Every one of them stops for a human before it commits.
Engineers who can start on your codebase, not just on your org chart. Short-term cover for a delivery crunch, permanent hires you keep, or a trial period before either side commits.
A distributed team that works your hours and sits in your standups. Scale it up for a release and back down afterwards without renegotiating everything.
One place where the numbers agree. We consolidate sources, model the warehouse, and set up access so finance, ops and engineering are reading the same figures.
Dashboards people open on a Monday morning. We build the metrics with the team that owns them, then the models that project them forward.
Trading and financial tooling, policy and claims systems, e-commerce and point-of-sale, connected-vehicle and manufacturing systems. Built to be handed over and maintained by your people.
A claims decision, a bound risk, a trade recommendation — none of these can come from a system that acts on its own and can't explain itself afterwards. Under the EU AI Act and DORA, that isn't a preference any more. We build agents that do the reading, show their sources, and wait for the person who is accountable.
The agent reads the claim, pulls the relevant policy wording, checks cover and flags the exclusions it found — then holds. The adjuster approves, edits or rejects. Every run leaves a record of what was read and who decided.
Retrieval across policy wordings, reinsurance treaties, prospectuses and regulatory filings that follows the structure of the document instead of chopping it into fragments. Answers point at the clause they came from.
Broker submissions arrive as email threads, spreadsheets and scanned PDFs. We extract the risk data into structured fields, flag what's missing, and route to the right underwriter. Nothing gets bound without a human.
Screening, exposure and research questions answered in the dashboard, not a separate chat window. Responses stream into real grids and charts, and the agent's tools update the view rather than describing what you should change.
Prices, fundamentals, volatility, claims history, exposure and derived signals — modelled into analytical stores and event streams. Most AI projects fail here, so this is usually where we start.
Every run records its inputs, retrieved sources, decision and approver, so the answer to "why did it do that" exists before anyone asks. Claude by default, other providers behind the same interface — model choice stays a configuration decision, not a rewrite.
A few days a month of senior direction: architecture calls, review, and saying no to the expensive ideas early.
Our engineers inside your process, your tools, your repo. You manage the work; we handle hiring and cover.
A warehouse, a dashboard, or a working agent — specified, built, documented and handed over to your team.
An audit of a stalled data platform or an AI proof-of-concept that won't reach production, with a written call on what to fix or kill.
The consulting practice works across all four. The AI services are built for the two where a decision has to be defensible.
Trading-adjacent tooling, risk and regulatory reporting, and the market data pipelines underneath them.
Claims, underwriting and policy administration — where volume is high, wording matters, and every decision needs a name against it.
E-commerce, point-of-sale and the reporting layer that ties stores, stock and online into one set of numbers.
Telemetry, diagnostics and manufacturing systems, where data volume is the hard part and downtime is expensive.
Send a couple of paragraphs about the system, the team and the deadline. You'll get a straight answer about whether we're the right people — including when we're not.