How to launch the preliminary diagnostics process in 4 months
A patient with chest pain visits the clinic website. They do not know whether emergency care is needed or they can make an appointment with a general practitioner. Patients lose time, doctors are overloaded, and the burden on emergency services grows.
Automated triage reduces waiting time by 2-3 times and cuts the number of erroneous emergency calls by 40%. We develop AI Symptom Checker systems that analyze symptoms in real time, determine urgency, and route to the right specialist.
Our experience includes more than 50 projects in medical AI, and we guarantee sensitivity for critical conditions of at least 100%.
A good symptom checker does not make a diagnosis — it helps make a decision: where to go and how urgently. It is the first point of contact, and its quality affects the burden on doctors and patient satisfaction.
According to a JMIR study, implementing a symptom checker reduces call center load by 30% and increases booking conversion by 1.5 times. The system pays for itself in 3 months, saving significant funds on call center expenses.
Clinical tasks of the system
- Triage by urgency: immediate emergency care, emergency department visit, scheduled visit, self-care, telemedicine.
- Differential diagnosis: list of probable conditions with probabilities.
- Referral to a specialist: eliminating unnecessary visits to a general practitioner.
How does the conversational interface and NLU work?
Modern symptom checkers use conversational AI, not checkbox questionnaires. NLU based on a fine-tuned medical LLM extracts symptoms from free-form speech, understands synonyms, asks clarifying questions, and accounts for negations and temporal characteristics.
Chat interface vs questionnaire: completion rate 73% vs 41% — 1.78 times higher. Patients are more willing to tell their story in a conversation format. Conversion to the target action grows by 30%, and request processing is 5 times faster than manual triage.
How do we ensure 100% sensitivity?
Safety-first design is the foundation of the system. We design to eliminate a false sense of safety:
- Never downgrade triage (if in doubt — a more urgent category).
- Explicit disclaimer: the system does not make a diagnosis, a doctor is mandatory.
- Red flags: any potentially serious symptom → immediately higher triage.
- Age and demographics are taken into account during triage (chest pain in a 55-year-old man vs a 20-year-old woman).
This reduces the risk of missing a critical condition. Models undergo independent audit.
Differential diagnosis model: Bayesian vs Neural
We compare two approaches: Bayesian networks and neural networks. Bayesian networks based on medical knowledge bases (symptom-disease matrices) + an ML component for adjustment to population epidemiology. Or an end-to-end neural network trained on real clinical cases.
| Characteristic | Bayesian network | Neural network |
|---|---|---|
| Interpretability | High (transparent probabilities) | Low (black box) |
| Sensitivity to rare diseases | Requires expert priors | Can learn if data is available |
| Ease of audit | Easy to verify | Requires additional tools |
Bayesian network is 2-3 times easier to validate and audit, so for safety-critical systems we recommend a hybrid: Bayesian+ML. Sources of knowledge bases: SNOMED CT, clinical guidelines.
Input data: symptoms (from the dialogue), demographics (age, sex), medical history (chronic diseases, medications), duration and dynamics of symptoms.
Development stages and what is included in the work
- Analytics and data collection — symptom labeling, knowledge base preparation.
- Model training and validation — LLM fine-tuning, Bayesian network setup.
- Integration and testing — REST API, chat interface, load testing (p99 latency < 200 ms).
- Deployment and support — MLOps setup (MLflow, Kubeflow), data drift monitoring.
Note what is included: API and architecture documentation (model card, data sheet), staff training, technical support for 3 months after launch, 99.9% uptime guarantee.
Approximate timelines
- MVP: from 4 months (basic triage, chat interface, 100 symptoms).
- Production: from 8 months (full differential diagnosis, EMR integration, validation).
- Cost is calculated individually after audit — depends on the volume of symptoms, required accuracy, and integration complexity.
| Stage | Duration | Result |
|---|---|---|
| Analytics and labeling | 1–1.5 months | Knowledge base, labeled symptoms |
| Model development | 2–3 months | LLM fine-tune, Bayesian network, metrics |
| Integration and testing | 1–2 months | REST API, chat, load testing |
| Deployment and support | 1 month | MLOps, monitoring, documentation |
Limitations and quality validation
Metric for a symptom checker: sensitivity for critical conditions should be close to 100%. Specificity is secondary. Validation is performed on real cases: we compare with doctors' diagnoses. Benchmark: Isabel DDx and Ada Health achieve 80-85% top-3 accuracy on standard cases.
Integration: mobile app, web widget, embedding into an EMR patient portal. Separate mode for healthcare professionals.
Contact us for cost and timeline calculation for your task. Request a demo version for your clinic.
