An idea, but no evidence yet
You have data and a use case, but do not yet know whether the promise is technically realistic or how to test it.
Common situations
When I can help
A model that is hard to defend
Performance looks promising, but robustness, bias, data leakage or practical limitations remain unclear.
A team that needs to decide quickly
You need an independent scientific opinion, a prototype or focused upskilling without immediately hiring a full-time specialist.
Engagements for healthtech, medtech and biotech companies, R&D teams, consultancies, training organizations and private healthcare providers.
Packaged services
A clear scope and a useful deliverable
Every engagement starts with a decision question. The required effort, data and success criteria are agreed before work begins.
Starting point
AI & data diagnostic
From €950
Typically 1–2 weeks
Quickly determine whether the project deserves further investment.
- Scoping interview
- Review of the need and available data
- Risks, technical options and success criteria
- Decision memo and prioritized roadmap
Core offer
Technical model audit
Typical budget: €3,500–€6,000
Typically 2–4 weeks
Evaluate an existing model before expansion, integration or presentation to partners.
- Protocol, metrics and reproducibility
- Data leakage, bias and robustness
- Error analysis and interpretability
- Actionable report and team debrief
Premium project
Prototype / proof of concept
Typical budget: €8,000–€15,000
Typically 4–8 weeks
Turn a priority use case into a measurable, documented demonstrator.
- Baseline and validation protocol
- Reproducible pipeline
- Iterations within a controlled scope
- Prototype, documentation and recommendations
Knowledge transfer
AI / LLM training for healthcare
Use cases, prompt design, limitations, validation, confidentiality and best practices adapted to professional roles.
Continuity
Scientific R&D advisory
Regular review of methodological choices, prioritization of experiments and preparation of key decisions.
Indicative budgets, adjusted to scope, data quality and security constraints. A precise quote is provided before every engagement.
Method
A simple, traceable collaboration
Audit and needs assessment
Discussion with your teams, available data, the decision to improve and definition of objectives.
Proposal and quote
Duration or fixed scope, exact price, number of iterations, deliverables, validation criteria and payment terms.
Delivery and handover
Progress reviews, validated deliverables, documentation, final debrief and transfer to your teams.
Evidence and demonstrations
Scientific expertise applied to concrete deliverables
Modeling health trajectories
Cohort structuring, meaningful target definition, out-of-sample validation and interpretation of factors associated with neurodegenerative trajectories.
Value: making a complex analysis defensible and useful for scientific decisions.Generative and predictive AI for faces
Generation, editing and analysis of visual data with experimental validation and peer-reviewed publications.
View the ACM publicationInteractive scientific dashboard
Transforming complex indicators into a bilingual exploratory interface for comparing scenarios and communicating results.
Confidential project — private presentation available on request.Engagement examples
Focused, measurable and realistic use cases
- Automated quality control for biomedical images: framing, artifacts and noisy data.
- Detection of incomplete records in a clinical information system using tabular data.
- Robustness, bias and data-leakage audit of a predictive model already used internally.
- Structured extraction of variables from medical or scientific PDF documents.
- Scientific dashboard for exploring cohorts, indicators or model results.
- LLM training for healthcare managers: useful applications, prompt design, limitations and GDPR.
Working framework
Security, validation and explicit limitations
Controlled data
Work on anonymized or synthetic data, or directly within the client’s controlled environment when required.
Documented validation
Baselines, data splits, appropriate metrics, error analysis and limitations recorded in the deliverables.
Responsible use
Prototypes support analysis and decision-making. They are neither certified medical devices nor clinical advice.
This consulting practice is independent and separate from my research position at École Normale Supérieure.
Next step
Start with a decision question, not a model.
Send me a short description of the context, available data, business or scientific objective and target timeline. I will recommend the most relevant engagement format.