Same data, different answer
Sampling is built into a language model. Ask twice, get two answers. A clinical decision can't rest on a coin flip that's invisible to the clinician.
0
Graded synthetic cases
Obstetric & gynecology clinical companion
Don't hand a clinical decision to a black box.
ObGynAssist™ is one clinical companion for obstetrics and gynecology — ten modules, from fetal-growth staging and twin assessment to ectopic workup, thromboprophylaxis and recurrent pregnancy loss — all built on a deterministic engine: the same data always yields the same answer, every threshold traces to a named guideline, and every decision can be audited rule by rule.
The problem
Why a language model isn't enough.
A general AI can sound right and still be wrong in ways you can't see. In obstetric and gynecologic clinical decision-making, three failure modes are disqualifying — any one of them on its own.
Thresholds that don't exist
Invented cut-offs, invented citations. Here the cost of a confident fabrication isn't a typo — it's a missed FGR or a mistimed delivery.
No way to ask "why?"
When a decision is questioned, you need to know which step fired which rule. A black box can't show its work — and these aren't tuning bugs, they're inherent.
The answer
Not intuition — a deterministic engine.
We didn't fine-tune a model. We built a rule engine that turns input into a sourced clinical plan through a fixed seven-step pipeline — and proves it every time.
A language model
- ✕Stochastic — output varies run to run
- ✕Can invent thresholds and sources
- ✕No audit trail — can't replay a decision
- ✕Reasoning is opaque and unversioned
- ✕Often needs the network and the cloud
The ObGynAssist engine
- ✓Deterministic — identical input, identical output
- ✓Every threshold wired to a named guideline
- ✓firedRules[] records exactly what fired
- ✓Stamped with engineVersion + rulesetSHA
- ✓Runs entirely on-device — no cloud call
Determinism
Not intuition. The same biometry and Doppler produce a byte-identical plan — yesterday, today, and five years from now.
Sources
Not guesswork. Each decision names the RCOG, NICE, ISUOG, ACOG or INTERGROWTH-21st reference behind it.
Traceability
Not opacity. When a plan is questioned, the audit trail shows precisely which step fired which rule.
One umbrella
Ten modules. One companion.
ObGynAssist spans the OB-GYN day: fetal growth and twins, eight pregnancy-care calculators, and gynecology — each module self-contained, each grounded in its own guideline, all sharing the same deterministic foundation.
FetalGrowth
FGR / SGA / AGA / LGA staging, Doppler analysis and eight twin pathways.
RCOG · ISUOG · NICE · DelphiPregnancy suite
Dating, ectopic, aspirin & VTE prophylaxis, thyroid, hyperemesis, ASA and CMV.
RCOG GTG · NICE NG · ISUOGRecurrent pregnancy loss
Evidence-based workup with APS- and PRISM-gated treatment logic.
ESHRE 2022 · RCOG GTG 17
The flagship · FetalGrowth
The deepest module: the whole spectrum of growth.
FetalGrowth is where the engine runs deepest — one module spans every size classification and both pregnancy types, and reads a trajectory rather than a single measurement.
FGR · SGA
Early- and late-onset growth restriction, staged by Doppler.
AGA
Appropriate-for-GA, with wellbeing and maternal overlays.
LGA · Macrosomia
Large-for-GA and macrosomia, diabetes-stratified.
Twin pairs
DCDA / MCDA / MCMA with TTTS, TAPS, sIUGR and discordance.
A trajectory, not a snapshot.
Crossing centiles, accelerating, stagnant, or deteriorating on Doppler — the engine evaluates how growth is moving, not just where it sits today.
8 pathways
4 centile standards
44 phenotypes
See its work
One case, falling through the engine.
No black box. Pick a case and watch it descend the real pipeline — single-field sanity and cross-field plausibility gates, an 80-way routing decision, 425 rules narrowed to the handful that actually fire, a 14-layer coherence pass, then a stamped, replayable plan. Same input, same descent, every time.
A faithful illustration of the real nine-stage pipeline (ClinicalPipeline.evaluate):
stage names, rule IDs and table sizes are the engine's own. Reduced to its skeleton for the web —
the shipping engine weighs many more inputs and modifiers. Not for clinical use.
The proof
The numbers behind the structure.
Measured on the flagship FetalGrowth engine — the same testing discipline backs every module in the app.
50,000 / 50,000
runs bit-identical across 50 configurations × 1,000 seeds — determinism, proven.
< 0.4 ms
p95 latency. A full sourced plan computes in real time, on the phone.
58 cards
guideline reference cards behind 425 rules — RCOG, NICE, ISUOG, ACOG, INTERGROWTH-21st.
Concordance is measured against an independent guideline oracle on synthetic cases — it reflects agreement with standard-of-care references, not neonatal outcomes. A prospective clinical-validation study is in design. Read the full validation →
The app
From first tap to a sourced plan.
The most correct engine changes nothing if it can't be used. ObGynAssist puts the engine in a calculator clinicians trust — enter biometry and Doppler, get a staged classification, an acuity label, surveillance timing and the citations behind every line. Shown below in the flagship FetalGrowth module; every module follows the same flow.
- ✓Real-time recompute as you type — 250 ms.
- ✓Non-directive labels — classification, not orders.
- ✓Patient data never leaves the iPhone.
Get started
Better than AI — because it has to be.
Not faster, not flashier. More reliable, more traceable, more correct. ObGynAssist is coming to the App Store as a Medical Reference companion. For early access or clinical partnership, get in touch.
Or email info@obgynassist.com
Coming to the App Store · Medical Reference