Project summary

I am developing a physics-informed, generative AI system for non-invasive blood pressure monitoring in animals, grounded in rigorous cardiovascular modeling and supervised by two Colorado State University (CSU) professors. This project directly addresses a major gap in veterinary medicine: the lack of accurate, affordable, and species-agnostic blood pressure tools for animals in both clinical and low-resource environments.

Current non-invasive BP devices for animals are notoriously unreliable, they fail under stress, motion, anesthesia, breed differences, and inter-species variability. Existing algorithms are often tuned to narrow physiological profiles, and collecting large real animal datasets is logistically impossible and ethically constrained.

To solve this, I am building a Sim-to-Real Transfer Learning Framework that can generate infinite amounts of synthetic, labeled hemodynamic data using:

• 4-element Windkessel cardiovascular models

• Allometric scaling for multi-species physiology (cats, dogs, horses)

• Behavioral and clinical event simulation (white-coat stress, anesthesia, pathology, motion artifacts)

• Nonlinear PPG synthesis with optical modeling of fur, melanin, and sensor noise

• A hybrid ML architecture (ResNet + Transformer + Domain-Adversarial learning)

These components collectively produce a high-fidelity “virtual veterinary patient” environment, enabling deep neural networks to learn physiological invariants, something impossible with current real-world datasets.

My final output will be:

• A validated synthetic hemodynamics simulator

• A cross-species generative PPG dataset (“VetSim-1M”)

• A robust BP estimation model capable of handling stress, motion, and species variability

• An open-source codebase for global veterinary use

• A Colorado State affiliated workshop or conference publication in 2026

This project integrates the advanced ML training I am receiving through the Algoverse AI Research Program, which strengthens the reliability, reasoning, and robustness of the modeling pipeline. I am supported by two professors who guide the biomedical and veterinary scientific validity, ensuring that the simulation and downstream models align with real physiology and clinical needs.

The outcome is a new paradigm for veterinary diagnostics: a low-cost, physics-grounded, AI-powered BP monitoring framework that can eventually help animals in third world countries and other low-resource regions, where invasive monitoring is unavailable and non-invasive devices routinely fail.

What are this project's goals? How will you achieve them?

The core objective of this project is to develop a scientifically grounded, non-invasive method for estimating arterial blood pressure in animals using a combination of physics-based simulation, generative modeling, and modern ML architectures. Accurate BP measurement is essential for diagnosing pain, cardiovascular disease, anesthesia safety, and stress responses in animals, yet existing devices are expensive, invasive, species-specific, or unreliable in non-clinical environments.

This problem is especially urgent in low-resource veterinary settings such as third-world countries and other Caribbean regions, where access to advanced monitoring tools is limited and animals often experience undetected hypertension, shock states, or distress due to lack of diagnostics. A low-cost, AI-enabled alternative would significantly improve veterinary outcomes and accessibility.

How will this funding be used?

This funding directly supports the technical, computational, and academic requirements of building a physics-informed, multi-species generative framework for non-invasive veterinary blood pressure estimation. Each cost corresponds to a necessary component of the research pipeline:

1. Advanced ML Training (Algoverse AI Research Program Tuition): $3,000

Algoverse AI Research Program provides the advanced machine learning education required for this project, including model reliability, reasoning frameworks, hallucination prevention, agentic evaluation techniques, and interpretability. These skills are fundamental for designing the hybrid ResNet + Transformer + DANN architecture used in the sim-to-real BP estimation model.

2. Travel to Colorado State University: $2,000

Round-trip travel from Trinidad to Colorado (December 2025) to meet with my two CSU faculty mentors and finalize:

• Simulation constraints and species-specific hemodynamic parameters

• Validation protocols for generated waveforms

• Model evaluation metrics

• Publication roadmap for 2026 workshops

This meeting is essential for aligning the veterinary, biomedical, and computational components of the project.

3. Dedicated GPU Hardware (DGX/RTX-Class System): $4,500

The research requires local access to high-performance GPU capabilities to train:

• Windkessel-informed generative waveform models

• Transformer backbones

• Domain-adversarial neural networks

• Multi-species calibration models

A DGX/RTX-class GPU system enables:

• accelerated training cycles

• rapid experimentation

• offline model tuning

• local reproducibility without relying solely on cloud instances

This hardware investment is crucial for simulation-heavy biomedical ML.

4. Cloud Compute (RunPods, AWS/GCP GPU Instances, API Credits): $1,000

Cloud resources complement local hardware for:

• Large-batch waveform generation

• Hyperparameter sweeps

• Running long training jobs

• Hosting the synthetic dataset (VetSim-1M)

• Paying for API usage in evaluation and experimentation

Funding ensures uninterrupted access to compute environments necessary for producing publishable results.

5. Research Operating Costs: $1000

This includes all software and scientific infrastructure required for running reproducible, open-access research:

Dataset Tooling

• Tools for generating, labeling, validating, and managing synthetic waveform datasets.

• Storage systems for high-volume time-series data.

Publication & Scientific Dissemination Fees

• Costs for submitting to ML4H, NeurIPS Workshops, AI4Science, or similar 2026 venues.

• Open-access fees if required for broader scientific reach.

Experiment Management & Logging Infrastructure

• WandB, Weights & Biases Pro, Neptune.ai, or related tools for tracking experiments.

• Model registry, versioning, and reproducible pipeline services.

Paid Libraries / Scientific Software

• Signal processing packages

• Physics simulation tools

• GPU-optimized libraries

• Dataset manipulation frameworks

• Latex templates (if needed for manuscript prep)

Quality + Compliance Tools

• Linters, testing suites, reproducible environment management

• Security/licensing tools required for open-source release

These operational tools are essential for producing a publishable, reproducible, and scientifically credible project.

Who is on your team? What's your track record on similar projects?

Professor Jesse Wilson: Principal Investigator(Colorado State University)
Professor Wilson oversees the scientific and technical direction of this project. He specializes in biomedical signal processing, optical sensing, computational physiology, and quantitative modeling of biological systems. As PI, he guides the design of the physics-informed simulation framework, model architecture development, and the publication strategy for 2026 workshop submissions.

Dr. Ellen Brennan-Pierce: Veterinary Physiology Mentor (Colorado State Veterinary Medicine & Biomedical Sciences)
Dr. Brennan-Pierce provides veterinary clinical context and ensures the biological realism of the simulation models. She advises on species-specific cardiovascular characteristics, behavioral and stress-related artifacts, and the clinical interpretation of synthetic waveforms, strengthening the accuracy and applicability of the project’s outputs.

Ashanna Narrie: Lead Student Researcher (Independent / Trinidad & Tobago)
I am the primary researcher developing the computational, simulation, and machine learning components under the guidance of the PI and mentoring faculty. I authored the project’s 17-page technical proposal, developed early ODE Windkessel prototypes, structured the multi-species modeling pipeline, and am responsible for implementing the simulation engine, training generative models, and preparing open-source tools and the 2026 academic manuscript.

Track Record
I previously conducted two years of research in Dr. Christopher Snow’s laboratory at Colorado State University (Chemical & Biological Engineering), working on biomolecular crystal growth and computational biophysics. This experience strengthened my skills in modeling, simulation workflows, data analysis, and scientific writing. Our work was presented at the Scott Undergraduate Research Symposium (2023). The current project originated after engagement with the Colorado State's Veterinary Teaching Hospital, where I observed firsthand the challenges associated with non-invasive animal blood pressure measurement. I subsequently developed a full technical proposal integrating multi-species cardiovascular modeling, PPG synthesis, and sim-to-real ML techniques. I am also completing advanced ML training through the Algoverse Research Program, focused on reliability, reasoning, and interpretability, all core skills required for this project’s AI components.

What are the most likely causes and outcomes if this project fails?

The most likely causes of project failure stem from technical limitations rather than organizational issues. One possible challenge is a mismatch between the synthetic signals generated by the simulator and the characteristics of real physiological data, which could make sim-to-real transfer less effective than anticipated. Species variability may also pose difficulties, as the cardiovascular dynamics of cats, dogs, and horses may differ more than expected, limiting the model’s ability to generalize across multiple species. Other potential points of failure include insufficient computational resources, which may restrict training of large Transformer or domain-adversarial models; instability in the ML architecture during cross-domain training; or limited availability of reliable veterinary physiological reference data needed for rigorous validation.

If the project does not fully achieve its intended goals, the outcome will still yield meaningful and scientifically valuable results. Even a partial success will produce a validated Windkessel-based hemodynamic simulator for at least one species, along with a functional PPG generative pipeline that can serve as groundwork for future veterinary ML research. The project will also generate early versions of the hybrid ML model that demonstrate feasibility, even if further refinement is required before achieving publishable accuracy. In all scenarios, the codebase, simulation framework, and methodological documentation will be open-sourced, allowing CSU students, Algoverse researchers, and veterinary engineers to build upon the foundation created. Thus, even if full cross-species BP estimation is not achieved, the project will still contribute durable research infrastructure and a clear roadmap for subsequent iterations.

How much money have you raised in the last 12 months, and from where?

I have not raised any direct funding for this project in the last 12 months. All early work, including drafting the technical proposal, building initial simulation prototypes, and coordinating with CSU faculty, has been self-funded. For the Algoverse Research Program, I was awarded a 30% tuition scholarship, but I have not secured any external financial support beyond that discount. This application represents my first request for dedicated funding to support the remaining Algoverse tuition, GPU compute, travel to meet with my CSU mentors, and the full development of this research into a 2026 publication.