Project Quintessence builds and trains medical AI models, validates them against clinical evidence, and integrates them into workflows where performance, uncertainty, and usability can be tested in practice.
Lab model
Our mission is to develop multimodal AI systems for medicine that remain clinically interpretable, evidence-aware, and honest about uncertainty. The goal is not only model performance, but model behavior that can survive expert review.
Focus
multimodal medicine, clinical reasoning, validation-stage AI
Standard
claims must match evidence
Outcome
research artifacts clinicians and institutions can examine
We work across clinical language models, agentic systems, forecasting, medical imaging, and physiologic world models. Each project is shaped around a clinical question, a technical method, and a validation path.
Tracks
language, agents, forecasting, imaging, physiology
Artifacts
models, benchmarks, studies, prototypes
Outcome
a connected research program, not isolated demos
Project Quintessence is currently research-first: some studies are completed, others are ongoing or in validation planning. The work is positioned before deployment language, where methods, evidence, uncertainty, and workflow fit can still be challenged.
Status
completed, ongoing, validation planning
Posture
research before deployment claims
Outcome
a clearer boundary between promise and clinical readiness
We scope each collaboration around the problem, the available data, the clinical workflow, and the evidence needed to evaluate the model. Development, validation, and usability are designed together rather than added after the fact.
Process
question, data, model, evaluation, workflow
Partners
hospitals, labs, clinicians, engineers, academic groups
Outcome
protocols, models, studies, and reproducible research outputs
Evidence strata
Each area contributes methods, outputs, active studies, and validation principles to the same evidence-grounded research system.
We focus our clinical AI research across five core disciplines—from medical language models to physiologic simulation. Each track is structured around a specific clinical question, a reproducible training methodology, and open-source evaluation tools so partners can inspect model behavior before deployment.
Research Domains
Clinical language models and NLP systems from data construction to expert-led validation.
Methods
Outputs
Validation principle
We evaluate models by clinical correctness, evidence traceability, calibration, robustness, and expert adjudication. Fluency is not treated as evidence of competence.
Projects
NephroGPT: A Tool-Augmented Large Language Model for Nephrology Training and Continuing Education
Specialist agent workflows with auditable tool use and evidence-grounded decision traces.
Methods
Outputs
Validation principle
Agentic performance is accepted only when plans, tool calls, evidence inputs, intermediate states, and failure boundaries are inspectable by clinical reviewers.
Projects
An Evidence-Grounded Multimodal Agentic AI Co-Pilot for Cardiologists
Risk estimation, temporal modelling, intervention analysis, and reproducible outcome validation.
Methods
Outputs
Validation principle
Prediction and forecasting models are evaluated by temporal holdout performance, calibration, drift sensitivity, intervention timing, and reproducibility across cohorts—not one-off fit metrics.
Projects
The Impact of COVID-19 Non-Pharmaceutical Interventions on Notifiable Infectious Diseases in Poland
Medical imaging models spanning chest radiography, report-derived supervision, and multimodal post-training.
Methods
Outputs
Validation principle
Medical vision models are evaluated against image findings, report agreement, calibration, subgroup robustness, and downstream clinical meaning; benchmark scores are not treated as sufficient validation.
Projects
Benchmarking Large Language Models for Automated CheXpert-Style Labeling of Chest X-Ray Reports
ChexQwen: A Post-Trained Multimodal LLM as a Foundation CXR Model
Learned models of patient-state dynamics that forecast trajectories and simulate counterfactual futures.
Methods
Outputs
Validation principle
A physiologic world model is valid only if its learned state dynamics preserve temporal coherence, cross-modal consistency, intervention sensitivity, and calibrated uncertainty against held-out clinical trajectories.
Projects
EchoWorld: Generative Echocardiography from ECG and Clinical Context