Explore Projects

I plan to make BRL-CAD python bindings. Basically my goal is to make a Python interface for BRL-CAD using MOOSE as the core modeling backend, with a thin C bridge between Python and C++ for stability and portability. The architecture is: Python interface -> moose_rt3 adapter (ctypes) -> C bridge -> MOOSE C++ API -> BRL-CAD database engine -> .g file. I plan to work around 3 goals: 1. Database lifecycle behavior should be consistent across read-only, writable and in memory modes with clear errors when operations are invalid 2. Core primitives and CSG operations should be exported correctly and transforms must be stored as combination-member matrices so results are verifiable by MGED. 3. Unified read access: provide one Python read entrypoint for supported object families with a stable return format and desterministic errors for missing or unsupported objects

This project aims to improve and evaluate symbolic regression methods for discovering interpretable mathematical models of biological network dynamics. A key challenge in this area is that, when working with real biological data, the true underlying equations are unknown, making it difficult to assess whether equation-discovery algorithms are accurate or biologically meaningful. To address this, I will generate synthetic time-series datasets from curated mechanistic models available in the BioModels database (e.g., SBML models), where the ground-truth equations are known. These datasets will be used to systematically evaluate and benchmark symbolic regression approaches, including PySR and neural-symbolic methods such as the Learning Law of Changes (LLC) framework. The project will also explore strategies to improve model identifiability and interpretability, including constrained search spaces and incorporation of biologically informed functional priors. The expected deliverables include: (1) a reproducible pipeline to generate simulation data from SBML models, (2) an evaluation framework to benchmark symbolic regression methods on equation recovery tasks, (3) implementation and comparison of PySR and LLC-based approaches, (4) strategies to address identifiability challenges in biological data, and (5) well-documented, open-source code integrated with existing SBML compatible tools.

This project aims to develop and integrate machine learning models capable of downscaling the carbon flux outputs from the Predictive Ecosystem Analyzer (PEcAn) framework to higher spatial and temporal resolutions. The coarse resolution of PEcAn's current outputs limits their applicability for localized analyses of carbon dynamics. By leveraging deep learning techniques, specifically Convolutional Neural Networks (CNNs) and Spatio-Temporal Recurrent Neural Networks (STRNNs), this project will enhance the accuracy and granularity of PEcAn's model outputs, enabling more detailed analyses to support ecosystem management and climate change mitigation efforts. The project will involve developing a robust data preparation pipeline to handle diverse covariate datasets, including climate variables, land cover, soil properties, and disturbance history. Time-varying covariates and disturbance events will be integrated to capture their influence on carbon flux dynamics. Tailored CNN and STRNN architectures will be designed and implemented to effectively capture spatial and temporal dependencies in the data. Comprehensive model training, evaluation, and optimization strategies will be employed to ensure accurate downscaling performance. The developed models and workflows will be seamlessly integrated into the existing PEcAn framework, ensuring accessibility and reproducibility for researchers and practitioners. User-friendly interfaces, command-line tools, and comprehensive documentation will facilitate the adoption and usage of the downscaling capabilities. The key deliverables include a scalable data preparation pipeline, optimized CNN and STRNN architectures for downscaling, model training and evaluation pipelines, evaluation metrics , and the integration of the developed solution into the PEcAn framework with user-friendly interfaces and documentation.

PEcAn's North American carbon reanalysis assimilates satellite and ground observations into SIPNET at 1km resolution, but has never been directly compared to TRENDY or CMIP model ensembles. This project builds a reproducible Python pipeline using ILAMB to score PEcAn alongside all available TRENDY and CMIP models on the same observational benchmarks across five to six variables (AGB, LAI, soil moisture, soil carbon, GPP, NEE). Deliverables include ILAMB scorecards, a reusable ensemble calibration module comparing within-model vs across-model uncertainty, regional diagnostic notebooks, and updated temporal benchmarks contributed back to ILAMB.

This project aims to integrate bytetrack algorithm into a Mediapipe graph using OpenVINO inference calculators. The bytetrack graph will be deployed on OpenVINO model server to perform efficient object tracking on a video input. The deliverables of this project include a bytetrack graph in OpenVINO mediapipe fork with custom calculators, YOLOv10 model and bytetrack graph packaged in OpenVINO model server, and a Python client for the bytetrack graph to perform bytetrack algorithm on video stream. The final phase of this project includes evaluation of bytetrack on metrics MOTA, IDF1, and HOTA. The goal of this project is to bring the bytetrack algorithm to OpenVINO model server to leverage its efficiency, scalability, and robustness for production-grade inference.

<p>Collaboration is ingrained into human nature, without it we, as a species, wouldn’t be able to build astonishing buildings like we have today. Fundamental to collaboration, however, is communication, which enables the coordination of big projects. To make the work on a project easier we use tools like FreeCAD. It offers possibilities to model objects, like buildings, but not exclusively, that shall soon be built. For that purpose FreeCAD supports BIM (Building Information Modeling). This project aims to integrate the BCF (BIM Collaboration Format), which is designed to communicate issues/topics with a particular model. These issues/topic can then be directly visualized in the design tool and don’t have to be searched for in the model, based on a description in a PDF file.</p>

The Computational Modeling in Biology Network (COMBINE) is an initiative to coordinate the development of various community standards and formats in systems biology and related fields. While multiple resources exist explaining these standards as well as libraries to work with these standards, it is currently challenging for newcomers to start using them. Existing resources are difficult to find, read, comprehend, implement, and interpret. The overall goal of the project is to address these issues by developing Jupyter notebooks showcasing COMBINE standards and libraries. The notebooks will be developed around one common theme; the famous repressilator model. The project will demonstrate how to construct, encode, visualize, simulate, annotate and distribute computational models using COMBINE standards and libraries. This project will contribute open resources which will ● introduce and educate newcomers to standards for computational models. ● aid in navigating the vast online resources available in this area by providing a common entry point. ● provide an interactive introduction to the libraries with code examples for important examples. ● provide testing of COMBINE libraries via Continuous Integration (CI). Thereby this project will add an important contribution to the field of systems biology and network biology.

This project aims to develop an OpenVINO GenAI extension for LangChain, enabling users to utilize OpenVINO as a backend for running language models. It involves integrating these components to ensure seamless compatibility, providing optimized performance for language model inference. Deliverables: 1. A fully functional OpenVINO GenAI extension for LangChain, allowing users to run language models with OpenVINO as the backend. 2. A finalized package distributed as a Python wheel.

<p>Open Geometry Viewer is a web-based application to view 3D CAD models on browsers. Users are able to host the model online and comment on others models. This project includes some bug fixes, enhancements and adds some new features to the existing application. This project will improve the end user's experience with the whole application. Along with that, this project includes some code quality improvements that will help other developers to understand and maintain the code easily in the future.</p>

The use of machine learning models in medicine is expanding, but there are still undergoing due to the interpretability. One useful example which has understandability is DrugCell, which reflects biological structures in machine learning models. Based on this model, this project aims to build a machine learning pipeline using DrugCell from the data in the CellMinerCDB database developed by NRNB. The current data structure of CellMinerCDB makes it challenging to pass inputs directly to DrugCell. Therefore, in this project, we will make PRs for 1) processing the data, 2) passing the data, and 3) calling DrugCell to make predictions. The deliverable of this project is to merge the above PRs into rcellminer, so that data acquisition, processing, and prediction can be completed by rcellminer.

<p>LibRoadrunner is a high performance SBML based simulator that uses LLVM to generate very efficient runtime code. This enables LibRoadrunner to simulate models on par with compiled C/C++ code. By combining Libroadrunner with standard optimization algorithms it is possible to use Libroadrunner to fit models to data. At present this is done by writing code to link the standard Python optimizer available via scipy with Libroadrunner. Although this works, it is inefficient and for large models, it is not practical. In this project, we would like to develop a C/C++ based differential evolution plugin that can be used directly by Libroadrunner (after integrating it with RRPlugins) without having to go via Python. This would enable us to provide high performance optimization capabilities.</p>

Currently, the Newt editor supports Systems Biological Graphical Notation (SBGN) and Simple Interactive Format (SIF) formats while also allowing importing/exporting data in Systems Biology Markup Language (SBML) format. Conversion to SBML format from SBGN or SIF is lossy since the data models are not fully compatible. Information may be lost during conversion to SBML format. This is why supporting the SBML format in Newt is crucial. The main feature proposed to be implemented is for users to create SBML models from scratch in Newt without the need to convert them to other data formats. An additional palette will be added with the corresponding notation to the existing palette for the user to create new SBML models or edit an existing one by dragging elements from the palette.

CySBML Reloaded: Smarter, Simpler, Automated focuses on upgrading CySBML — a Cytoscape app for visualizing computational models encoded in SBML. The project will enhance usability, improve stability, and introduce automation features, making it easier for researchers to explore, validate, and interact with complex biomedical models through an intuitive and streamlined interface based on network visualization. The main project outcomes of CySBML Reloaded: Smarter, Simpler, Automated will be a major release of CySBML, with an updated backend supporting automation features, an up-to-date frontend addressing usability issues, updated REST APIs, improved reading of SBML, and improved visualization capabilities such as network and cofactor collapse. This will allow us to integrate CySBML in modeling workflows based on Python/R.

<p>LibRoadrunner is a high-performance SBML based simulator that uses LLVM to generate very efficient runtime code. This enables LibRoadrunner to simulate models on par with compiled C/C++ code. By combining Libroadrunner with standard optimization algorithms it is possible to use Libroadrunner to fit models to data. At present this is done by writing code to link the standard Python optimizers available via scipy with libroadrunner. Although this works it is inefficient and for large models, it is not practical. In this project, we would like to continue the work to develop a C/C++ based optimization plugins that can be used directly by LibRoadrunner with the help of RRPlugins. RRPlugins is a software that uses Libroadrunner as a dependency to load plugins and perform optimization tasks. Now if we build roadrunner we need to build plugins again since plugins are designed in such a way that they use LibRoadRunner and RRPlugins both as a dependency. To overcome this we decided to integrate both LibRoadRunner and RRPlugins and this would enable us to provide high-performance optimization capabilities.</p>

The main goal of this project is to create a PyTorch Geometric(PyG) dataset by integrating the cBioPortal and Pathway Commons datasets and provide a clear example code for this integration. The project will require working with different datasets from the cBioPortal and Pathway Commons. The proposed solution involves a three-stage process: retrieving and preprocessing data from the datasets, integrating the data, and developing and training Graph Neural Network(GNN) models on the integrated dataset. The main deliverables for this project will be: 1. Preprocessed and integrated example dataset for PyTorch Geometric using cBioPortal and Pathway Commons datasets. 2. Example code for combining cBioPortal and Pathway Commons datasets in PyTorch Geometric. 3. Developed and optimized GNN models for downstream tasks using the integrated dataset. Additionally, the project will produce a well-documented and well-structured codebase for data retrieval, preprocessing, integration, and GNN model development. The final deliverable will be the contributed PyTorch Geometric dataset to the library for potential use by other researchers.

This project aims to develop an agentic AI system for predictive maintenance of urban infrastructure using OpenVINO. The goal is to enable intelligent monitoring and decision-making by combining multimodal perception with reasoning capabilities on edge devices. The proposed solution integrates a vision-language model to detect defects (Deformation, Obstacle, Rupture, Disconnect, Misalignment, Deposition) from visual data and a language model to interpret these defects in terms of severity, risk, and maintenance priority. These components will be connected through an agentic pipeline that performs perception, reasoning, and action in a unified workflow. The system will be implemented using OpenVINO-optimized models and deployed on Intel edge hardware to ensure efficient, low-latency inference. Open datasets will be used to validate the system in realistic scenarios. Expected deliverables include: - An end-to-end agentic AI pipeline for predictive maintenance - Integration of multimodal perception and LLM-based reasoning modules - Deployment and optimization using OpenVINO on edge platforms - Benchmark results evaluating performance and system effectiveness - Comprehensive documentation and reproducible codebase

When tackling the challenge of detecting defects in high-resolution images, we encounter many challenges. One significant problem is the difficulty in training models effectively, primarily due to memory limitations. The process of down-sampling these images to fit within memory capacities leads to the loss of crucial information, which can result in the oversight of small yet significant defects. To overcome these issues, we can employ a strategy involving a tiled ensemble mechanism. In this case, images are split into smaller, manageable tiles. Each of these tiles is then processed using a separate model, specifically trained for that tile position. This strategy enhances localization and improves the detection of small anomalies, while also satisfying memory limitations. Anomalib currently has a basic tiler. However, the advantages of locally-aware models, like Padim, are diminished without the support of a tiled ensemble. To address this problem, we introduce a tiled ensemble implementation that preserves the advantages of local awareness while also enhancing overall performance. This approach aims to incorporate Anomalib's existing features, while simultaneously providing users the flexibility to tailor specific steps within our pipeline. All the while, we ensure that our approach remains memory-efficient.

This proposal is provided by OpenVINO team. I just copy and paste here. Short description: Automatic1111 is a powerful web user interface based on Gradio library specifically designed for Stable Diffusion. It’s most popular open-source Stable Diffusion WebUI on GitHub with 119K+ stars, which supports a lot of features like text-to-image, image-to-image, inpainting, Lora models, custom models from model hubs like civitai.com and huggingface etc. OpenVINO support for Automatic1111 enables Stable Diffusion run on Intel CPUs and GPUs, this solution is currently supported using a custom script. Implementing OpenVINO through the Automatic1111 extension will provide an easier way to use OpenVINO. This project will also aim to provide support for more AUTOMATIC111 features with OpenVINO. Task Description: Develop a built-in extension for Automatic1111 SD WebUI based on the existing OpenVINO custom script leveraging Diffusers library. Support some of the new features like Hires upscalers, new samplers, tiling, face restoration etc. Develop test scripts to evaluate these features. Evaluate with different Stable Diffusion variants (V1.5, V2.1, XL, LCM etc.) on Intel CPUs and GPUs. Optional: Evaluate compatibility with other extensions like ControlNet. Expected outcomes: Raise a PR with all the contributions to the OpenVINO fork and eventually to the mainstream Automatic1111 repo. Documentation with clear description of features in the extension and demo videos. Medium/OpenVINO blogs.

In this project, I plan to extend PEcAn by building two data ingestion workflows that will improve the modelling capabilities of PEcAn. The first will ingest weather forecast data from ECMWF, and the other will adopt a standard for agricultural management practices such as planting and fertilisation that are required by crop models.

<p>A wide range of research and engineering applications involve computing the thermodynamic properties of substances that are in the incompressible regime. I propose to append a model for incompressible fluids to the existing models in FPROPS. The simplifications in the formulation of incompressible EOS can significantly increase the performance of the computations compared to those with the general Helmholtz based EOS.</p>

<p>ModelBricks is a tool for bioinformatics researchers that allows users to browse, view, and download detailed models of biological processes. I propose that various needed visualizations, and model retrieval functions are added using the existing code framework and the VCell API. Furthermore, I propose that the VCell API be expanded to include useful functionality such as more versatile searching and permissions changes.</p>

<p>There is a need for creation of such a website that is Python Extensible as the new models are needed to be added for plotting which are not supported in Java so there is a need to create such a website through which we can visualize the plots that are generated and Python as a software tool is required to visualize those outputs of models which are the plots of the required extension which are in the form of high resolution plots. In this process there is a need to remove the whole dependency of Java from the website and convert that to the Python based format for accessing them in the website and to support publication quality plots with high resolution.</p>

<p>In PEcAn, there are data processing pipelines that prepare data as model drivers, as well as constraints. Flux tower networks are one of the most important sources that collect such data. The aim of this project is to add the Integrated Carbon Observation System(ICOS) data stream to PEcAn. This would provide a wide variety of input data for running the ecosystem models, as well as constraining them in the European environment. Additionally, this project aims to add a GEDI LIDAR data stream to the remote data module.</p>