Founder-led advisory · Finland
I turn data and AI into decisions that hold up: in the business case, in the workflow, and in front of whoever asks the hard question later. Increasingly that means agentic AI systems, built on frontier models or on open models that never leave your building. Life sciences is where my history is deepest. It is not where the work stops.
Agentic AI · Data science · Governance · Evidence | 10+ years leading this work inside complex organisations
Four practices. Most engagements start in one and pull in the next. The first three travel anywhere; the fourth is for organisations carrying a regulatory burden.
Working systems, not slideware: agents that retrieve, synthesise and act under human approval, on frontier models from any of the major labs or on open-weight models running on your own hardware. Traceable by design, and measured against your own experts rather than a benchmark.
The unglamorous half that decides whether anything else works. Getting the data into a state worth modelling, choosing methods that match the question, and reporting results with the uncertainty left in rather than rounded away.
Classify your AI use cases, decide what genuinely needs a control and what does not, and write governance people can follow. Where regulation applies, EU AI Act, GDPR and GxP get read together rather than as three separate projects.
For the life-science and medical-device end of the spectrum: evidence that answers the question a payer or regulator will actually ask, and a quality system built as living documents rather than a binder nobody opens.
Agentic AI
An agent here is a working system, not a chat window. It plans a task, calls the tools it needs, checks its own output, and hands the consequential decisions to a named person. I design and build these end to end, on whichever models the job and the data allow.
The practice itself runs on the same tooling: agentic coding assistants, a searchable knowledge base built from every past project, and skills that hold the working methods. It is how a one-person firm ships and maintains production systems.
Small, senior and direct. You work with me and a short list of trusted partners. No junior layers, no handovers, no dilution.
We work through your problem, your constraints and what "done" would look like. You leave with a written read of the situation whether or not we continue.
A defined deliverable: a governance position, a working agentic system, an analysis, an evidence plan, an audit-ready document set. Scope and pace are set by the problem, not by a standard package.
Retained advisory for teams that want a senior second opinion on call rather than another full-time hire, and for systems that need someone who knows why they were built the way they were.
Seed-stage startups through to global pharma, and organisations with no life-science connection at all. Client names withheld by agreement; happy to talk any of these through in a call.
Global pharma · R&D
Designed and delivered an agentic evidence-synthesis system for an R&D function: parallel retrieval across scientific databases and regulatory sources, human approval gates before anything reaches a reader, a full audit trail, and model routing that puts a fast model on screening and a reasoning model on compliance.
Global pharma · Quality
A regulatory-intelligence agent for a quality function. It watches guidance and regulation as they change, maps each change onto the procedures it touches, and drafts the impact assessment for a named owner to review and sign.
Pre-clinical biotech
Target and asset intelligence assembled by agents working over public literature, patent and structural data through purpose-built MCP servers, so a small scientific team could triage candidates without hiring an informatics function.
Digital therapeutics
Regulatory pathway, quality system and evidence plan for a digital therapeutic, taken together so the clinical claim and the software classification stayed consistent.
Consumer research · local only
A concept-testing panel that runs entirely on open-weight models on one workstation, with a fast tier and a careful tier the user chooses between. No data leaves the machine and there is no per-call cost.
Outside life sciences
AI use-case triage and a governance position for an organisation with no GxP burden, where the real constraints were procurement, public accountability and an ageing data estate.
I have spent more than a decade leading data, AI and digital transformation inside health and life sciences, long enough to have shipped things that worked and to have watched good ideas die in governance. That is where the deepest history sits: startups, pre-clinical and clinical biotech, DTx, digital health and global pharma. It is not a boundary. The same data science and AI work travels to industrial, public-sector and other commercial settings, and the questions turn out to be the same three: how to use data and AI responsibly, how to build capability that outlasts the project, and how to translate digital into business.
Day to day I work with frontier models from every major lab and with open-weight models on my own hardware, and I switch as the field moves. The range is deliberate. The right answer for a global pharma function and the right answer for a team whose data cannot leave a laptop are rarely the same model, and the client should not have to care.
The approach is Nordic in the practical sense. Design for the person doing the work, be honest about uncertainty, and prefer a smaller thing that runs to a larger thing that is still being aligned.
If you want clarity, acceleration, or a partner for the next chapter of your digital work, write to me directly. I answer my own mail.
PalsaIQ · Profile Sammeli Liikkanen Oy · Company ID FI3106765-4 · Finland
