The company.
IVF-DM (In Vitro Fertilization Decision Making) builds decision support for reproductive specialists. At each decision of an IVF cycle it estimates the odds of every option, with an honest range and the reasons behind it. The doctor decides.
- Stage
- Research stageDesign complete, models not yet trained
- Founded
- 2026Concept designed in 2020
- Based in
- Haifa, Israel
- Team
- Two foundersMedicine and engineering
- Funding
- Self-fundedNo outside investors
- Legal
- Operated by Konstantin TikhaevRegistered business in Israel
Product
A cycle turns on a few decisions. IVF-DM gives each one its own set of models.
- How to stimulatedrugs, starting dose, adjustments
- When to collecttrigger type, retrieval timing
- How to fertilizeIVF or ICSI, culture conditions
- What to transferfresh or frozen, which embryo, how many
For every option the doctor sees the estimated chance of success, of failure and of complications, the uncertainty around each estimate, and the factors that raise or lower it for this patient.
Customers: IVF clinics and reproductive specialists, starting in Israel
Data and privacy
HFEA anonymised register
UK treatment cycles from 1991 to 2018, identifiers removed, available to researchers on application.
Holds: outcomes, IVF or ICSI, fresh or frozen, embryos transferred, patient history.
Lacks: drug doses, AMH, follicle counts, trigger day, culture media, embryo grading.
The clinic's own records
What the register lacks, used only under the clinic's Helsinki (ethics) committee approval. We are looking for a partner clinic.
Rule: no identifiable record leaves the clinic. Models learn from de-identified records, inside the clinic.
Technology
- ModelsMulti-class gradient boosting and Bayesian models, one set per decision point. The Bayesian part gives an honest range, not just one number.
- Phase 1Training and validation on the public register on GPUs: RAPIDS cuDF, XGBoost on CUDA, Bayesian models in JAX, per-patient explanations with GPU TreeSHAP.
- Phase 2A GPU node inside each partner clinic running NVIDIA FLARE for federated XGBoost, so records stay on site.
- In the clinicConnects to the clinic's records system through an API, or works on its own with manual entry.
Team
Dr. Kseniia Tikhaeva
Founder and medical lead- MD, PhD. Obstetrician-gynecologist, more than 12 years in medicine, more than 7 in IVF.
- Co-developed a medical records system for IVF clinics (2016 to 2017).
- Publications and a patent in reproductive medicine.
Konstantin Tikhaev
Co-founder, technology lead and developer- Information systems engineer who builds and runs AI systems in production.
- Built A³, a real-time talking-avatar agent running on NVIDIA GPUs (US provisional patent).
- Runs an open study of how AI models describe themselves.
Roadmap
- 2016Kseniia co-develops records software for IVF clinics.
- 2020IVF-DM designed: decision points, modelling approach, a three-stage research plan.
- 2026Relaunch in Israel. Research stage, self-funded. We are here.
- NextAccess to the public register and first models; a clinician prototype, target 2027; a pilot with a partner clinic, then federated training across clinics.