Founder and CEO of TCR App Mobility — a Polish deep tech startup from the area of precision medicine and road traffic safety (website, pitch deck, video presentation). Seasoned machine learning engineer with extensive experience in physics, mathematics and data analysis. Award-winning architect of deep neural networks and multi-modal probabilistic algorithms with practical market application.

 

Over 9 years of experience in the development and commercialization of innovative solutions, gained during collaboration with stakeholders from 17 countries on 4 continents. Over 7 years of experience in the field of artificial intelligence (e.g., neural networks, deep learning, data mining), proven by successful validation and real-world implementation of advanced mathematical models.

 

Finalist of prestigious (domestic and international) startup competitions, conventions and festivals. Participant of distinguished business acceleration programs (e.g., Creative Destruction Lab AI Stream, INSEAD AI Venture Lab, Polish Development Fund's School of Pioneers). Exhibitor at more than 20 global tech conferences (e.g., MWC Barcelona, GITEX Dubai, South Summit Madrid, Salon des Inventions Geneva).

​​​​​

RESEARCH

​​

ML-based Prediction of Road Traffic Accidents Caused by Tiredness and Inattention (2021–2023)

​

The mathematical models of TCR App Mobility are rooted in two years of rigorous scientific research with novel findings about human cognition. Participants completed a multi-week longitudinal assessment phase via our SaMD-compliant mobile application, which preceded sessions in the custom-engineered VR driving simulator synchronized with EEG, fNIRS, eye tracking (saccadic latency, fixation precision, microsaccade dynamics), sEMG, GSR, pressure mapping and vehicle telemetry data. This workflow allowed us to precisely correlate actual brain activity with entangled driving dynamics and subconscious behavioural patterns manifested while operating a vehicle. Our proprietary dataset (one of the largest of its kind globally) includes digital health phenotypes of individuals from diverse demographic backgrounds, encompassing a broad range of driving attitudes and ageing profiles.

​​

TCR App Mobility is implemented as an autonomous mobile application (agentic AI) with a proprietary data flywheel. Our solution emits sound alerts whenever a dangerous data pattern is detected — averting the natural lapses in attention at least three minutes before any discernible cognitive impairment occurs. It ensures sustained awareness throughout the entire journey.

 

Presymptomatic Medical Diagnostics on the Basis of Subtle Driving Anomalies (since 2025)

​

By integrating a VR-based training workflow with diagnostic-grade feature engineering and our original approach to adversarial inverse reinforcement learning, we have translated complex neurophysiological impairments into measurable drifts of AIRL-derived driving reward functions in our infinite-dimensional feature space associated with a positive-definite kernel (referred to as the "diagnostic space"). It enables us to detect neurodegenerative diseases (as well as confounding non-CNS conditions such as diabetes, rheumatoid arthritis and glaucoma) 4 to 8 years before the onset of clinical symptoms.

​

Simultaneously, our models can uncover high-entropy anomalies that are associated with early-stage brain tumours, effectively allowing intervention when they are still fully curable.

​​​

In the case of Parkinson's disease (PD), the process we identify is not the appearance of resting tremors — symptoms which will remain imperceptible for another 5–10 years due to cerebral homeostatic plasticity. Instead, our AIRL-derived driving reward functions provide a mathematical representation of how the patient's brain allocates its resources at the most fundamental level (e.g., controlling lane position, managing speed and reacting to external stimuli while masking early-stage neurodegenerative deficits through compensatory axonal sprouting). Even subclinical micro-damage to dopaminergic neural pathways manifests as a discernible signature in the longitudinal evolution of these proprietary analytical constructs (referred to as the "driving phenotypes") in our diagnostic space, allowing us to capture the functional impact of the first 5–10% of neuronal degeneration.

​

We utilize proprietary neural network architectures (a purpose-built LSTM and a Temporal Geometry-Aware Transformer in two distinct computational paradigms), an original adversarial inverse reinforcement learning framework, a self-supervised disease identification pipeline, and empirically-derived topological constructs developed through two years of rigorous scientific research. Our approach ensures objective diagnostic accuracy that is immune to demographic variances in driving attitudes and ageing traits.

​

TCR App Diagnostics embeds a clinical-grade tool in a consumer-grade device. We are turning the world's road infrastructure into a neurological screening lab — identifying Parkinson's disease with hospital-grade sensitivity (>92%) and specificity (estimated 94%), approximately 5 years earlier than advanced clinical brain scans.

 

Personalization of Foreign Language Learning Materials on the Basis of Eye Behaviour and Facial Expression Analysis (2023–2024)

​

Our proprietary technology introduces a paradigm shift in computer-assisted language learning (CALL). Traditional educational platforms rely on performance-based heuristics to adjust content difficulty. In contrast, our system deploys a multimodal behavioural biometrics framework that dynamically adapts foreign language curricula to the unique memory encoding patterns of the individual learner. By leveraging consumer-grade hardware — such as smartphone and laptop cameras — our system actively monitors high-frequency physiological indicators, including ocular dynamics (gaze trajectory and directional vectors), pupillometric fluctuations, blink rate variability and microscopic facial expressions. Through real-time computer vision and deep learning pipelines, these behavioural data streams are synthesized to quantify the user's immediate cognitive load, engagement matrices and information retention thresholds. By identifying the precise chronological windows and contextual formats where optimal memory consolidation occurs, our technology maximizes the efficiency of long-term knowledge acquisition. In practical application, this engine functions as a dynamic, context-aware content generator (GenAI).

​

To process volatile behavioural biometrics in real time, our core model utilizes a hybrid convolutional neural network (CNN) and Transformer architecture. This multi-modal framework is designed to simultaneously ingest and align disparate data streams. The frontend consists of a lightweight CNN that processes the raw video feed from standard user cameras. It performs real-time facial landmark detection, tracking microscopic facial expressions and ocular dynamics (gaze vectors, pupil diameter and blink rates) with minimal computational overhead. The extracted spatial features are then projected into a temporal embedding space and fed into a Transformer network. Utilizing self-attention mechanisms, the Transformer analyses the sequential dependencies of these biometrics over time, decoding the latent relationship between the user's subconscious physiological states and their immediate cognitive load.

​

The empirical data reveals a transformative acceleration in knowledge acquisition and retention. The experimental cohort utilizing our platform achieved the target proficiency threshold on average 35% faster than the control group practising via traditional methodologies (both classes exhibited identical initial language level and maintained an equivalent total time commitment throughout the evaluation lifecycle). This statistically significant reduction in time-to-proficiency is directly attributed to the system's ability to prevent cognitive overload. By modulating lexical density and syntactic distribution at the exact moment the biometric pipeline flags fading retention, our algorithm ensures that every minute of user engagement occurs within the optimal window for memory encoding.

​

NOTABLE DISTINCTIONS

​​​

With TCR App Mobility

​

Acceptance into Creative Destruction Lab AI Stream (San Sebastian, 2025/26)​

​

Graduation from INSEAD AI Venture Lab (2025)

​

Top 10 global startups at the 6th Brain Innovation Days by European Brain Council (Brussels, 2026)

​

Top 16 Polish startups (including Top 3 Polish startups from the sector of modern economy) at the 16th European Economic Congress (2024)

​

Top 20 European startups at Infoshare Contest (2024)

​

Classification among Deep Tech Pioneers by Hello Tomorrow (2024 and 2025)

​

Top 20 startups at Mobileheroes Global (2024 and 2025)

​

Top 25 startups at OIST Innovation Accelerator (2024)

 

With other startups

​

Top 16 Polish startups (including Top 4 Polish startups from the sector of business processes) at the 17th European Economic Congress (2025)

​

Winner of the 7th Edition of the Polish Development Fund's School of Pioneers (2024)​

​

EXPERTISE AND RESPONSIBILITIES

​​​

Artificial intelligence

​

Deep neural networks (e.g., RNN, LSTM, CNN, PNN, Transformers), machine learning (supervised, unsupervised, self-supervised, reinforcement), multi-modal data fusion, generative AI, RAG, MCP, fine-tuning, LLMs, agentic AI

​

Data science

​

Data mining, statistics, data processing, cloud computing (e.g., AWS, GCP), data structures, data visualization, data transmission, data architecture design​

​

Software development

​

C/C++, Python (e.g., PyTorch, TensorFlow, Matplotlib, NumPy, SciPy, scikit-learn), mobile applications (e.g., Flutter), web platforms (e.g., Django), workflow automation and deployment

​

Mathematical modelling

​

Algorithm development, analysis, testing and review; infinite-dimensional topology

​

Scientific research​

​

Research planning, cohort selection, data analysis, documentation

​

Business

​

Startup leadership and management, business presentation, technology validation, marketing, piloting, customer acquisition, venture incubation/acceleration, fundraising