Founder-led consultancy
I help medical communications, pharmaceutical and biomedical technology teams identify, test and implement responsible automation for high-stakes document workflows.
Solutions are designed around human review, document fidelity and the option of fully local or client-controlled deployment when sensitive material cannot leave the organisation.
The difference
Know which steps can be automated, which require expert control and where errors would fail quietly.
Build human review, traceability and failure checks into the workflow so plausible output is not mistaken for reliable output.
Design around data policy, infrastructure, model quality and the people responsible for the final output.
Services
The first question is where AI belongs in the process—and where it does not.
01 / ASSESS
Map the document lifecycle end to end and identify what can be automated safely, what should remain human-led and what could fail without being noticed.
Output: workflow map, risk analysis and prioritised pilot recommendations.
02 / BUILD
Prototype or build tools for a defined internal workflow, including fully local deployment where policy prevents client material from reaching third-party APIs.
Designed for the workflow rather than adapted from a generic assistant.
03 / ENABLE
Practical training for writers and editors: where AI helps, where it can silently degrade quality and how to evaluate output rigorously.
Built around real documents, review responsibilities and team policy.
04 / ADVISE
Product direction, workflow validation and medical-content expertise for AI and health-tech teams building tools for pharma or medical communications.
Useful when the technical team does not have medcomms experience in-house.
How an engagement starts
Start by establishing whether the workflow, evidence and constraints support responsible automation—and what the smallest useful next step should be.
A 20-minute conversation about the workflow, problem and practical constraints.
A paid, focused review producing risks, priorities and a defensible next step.
A prototype, implementation, training programme or ongoing advisory support where warranted.
Selected work
Word-native manuscript review automation
A medical writer reviews an LLM-drafted revision and reply for each comment, then exports a real Word file with tracked changes, threaded replies, formatting, tables, images and citation fields preserved.
The custom OOXML parser and serializer can run with a local model via Ollama or an approved OpenAI configuration.
Read the Stet case studyData extraction, screening and evaluation
A tool processed more than 1,800 abstracts on multivitamin supplementation and extracted structured fields at machine speed.
It also misclassified context in ways that could corrupt a review. The workable design was hybrid: use the model for first-pass volume while a human expert retains interpretive control.
Read the published analysis (opens in a new tab)About
Medical communications specialist and software builder.
I spent seven years writing and reviewing medical content inside established agencies and publishers. I now apply that domain knowledge to automation for biomedical documents.
Experience since 2016 developing and reviewing peer-reviewed publications, congress communications and medical affairs materials for international pharmaceutical teams.
GPP, ICMJE/Vancouver, PubsHub and EndNote, with hands-on experience of author and reviewer workflows.
Python, OOXML, local LLM deployment, tool-calling agents, FastAPI, HTMX, pywebview and document extraction across common biomedical file formats.
BSc (Hons) Medical Microbiology & Immunology; PGDip Bioscience Enterprise; three years of medical training.
Start with the workflow
Describe the documents, the people involved and the constraint that makes the current process difficult. The first step is a short fit call, not a commitment to build software.
Working hours: CET · International B2B engagements