Explore Projects

<p>Energy system components are modular in nature and interact across multiple scales. There is a need for a modelling framework that can easily construct and configure systems of modular entities (energy demands, sources, converters, storages and network links). These systems should be hierarchically nestable, facilitating the construction of large systems at high levels of detail. The framework should also facilitate the approximation of sub-systems, to aid model solvability. The above framework should be implemented in an object-oriented fashion, making use of the modular &gt;&gt;Open Energy System Modelling Framework&lt;&lt; (Oemof). Components should be specified in a common format that integrates with a database of available entities. The implementation should allow particular systems to be duplicated and reconfigured as well as reused as subsystems in larger models. Configuration parameters of multiple sub-systems should be readable from an external table.</p>

VLA models like unifolm-vla combine a vision-language backbone with a flow matching action head. On Intel iGPU, these models face two major deployment hurdles. First, structural blockers in the model code prevent successful OpenVINO conversion. Second, the iterative denoising loop requires kernel-level optimization to reach the frequencies needed for real-time robotic control. I will resolve four specific export blockers related to Python loops, random number generation, and precision handling. After achieving a successful export, I will implement kernel fusions for AdaLayerNorm and optimize Scaled Dot Product Attention to reduce latency on Arrow Lake and Panther Lake hardware. Project Deliverables: - A working OpenVINO export pipeline for unifolm-vla with all conversion blockers resolved. - Kernel-level optimizations including AdaLayerNorm fusion and SDPA pattern fixes. - A reference project featuring export scripts, quantization configurations, and latency benchmarks. - A detailed profiling report identifying bottlenecks and primitive selection via VTune.

This project proposes the development of a Python web application that utilizes Large Language Models (LLMs) for extracting, transforming, and visualizing molecular interaction networks from scientific literature. Leveraging LLMs' advanced natural language processing capabilities, the application will interpret complex academic texts to construct interaction graphs. These results would be uploaded to the Network Data Exchange (NDEx) platform, enabling researchers to manage and visualize biological data more effectively. Key deliverables include: An interactive Python web app for document processing and network generation. Seamless integration with the NDEx API for data management and graph visualization using Cytoscape. A robust pipeline for document analysis and data transformation with CX2 format output.

<h5>Checker Creation Help Script</h5> <p>Since, cve-bin-tool relies on contributors to build checkers for each package. This in turn requires them to know various file manipulation methods, which are easier to use on linux than on windows. To simplify this process, I propose to create a checker creation helper script, which would take .rpm, .deb, .tar.gz or other files that represent packaged versions of the software to be detected, including the product name and version number that we expect to find in each as a set of inputs. The script would automate the process mentioned in the <a href="https://github.com/intel/cve-bin-tool/blob/main/cve_bin_tool/checkers/README.md" target="_blank">checkers/readme.md</a></p> <p><strong>Recommending Safe Package Versions</strong> - Currently, cve-bin-tool outputs the vulnerable package name, version and vendor-product name. An upgrade to this would be to also provide the user with the list of all safe/fixed packages for the corresponding vulnerable packages found while scanning. This could be done by taking a vulnerable package (which was found in the scan) and web-scraping for the latest vulnerable version and then comparing it with all it’s released versions and then giving the safe version as output. An extended goal to this could be to automate the final downloading process.</p>

<p>ModelPolisher is a model annotation tool for the BiGG Models Knowledgebase. Annotations enhance the reusability and interoperability of biological models. This project will improve ModelPolisher by extending its annotation capabilities for models lacking BiGG Identifiers. Also, option to produce separate glossary file and embedding them in COMBINE archive will be added. Further ModelPolisher will be containerized to simplify setup of its database back-end and the project will be updated to software development standards.</p>

This project aims to fine-tune Vision-Language Models (VLMs) for domain-specific tasks like object detection and hierarchical classification, focusing on real-world applications in fields such as medical diagnostics and biodiversity monitoring. Although VLMs like BLIP-2 and TinyCLIP show strong general-purpose performance, their effectiveness diminishes in specialized domains due to domain shifts, hierarchical label structures, and hardware constraints. To address this, we will implement efficient fine-tuning strategies—including LoRA, QLoRA, and adapter tuning—within the OpenVINO Training Extensions (OTX), enabling scalable training on consumer-grade GPUs. We’ll enhance these models with support for structured prediction and deploy them using OpenVINO’s quantization and optimization toolchain, ensuring fast, lightweight inference across devices. Beyond technical contributions, the project delivers practical tools and open-source pipelines that bring advanced multimodal AI to resource-limited environments. By supporting transparent, efficient deployment of VLMs in impactful domains, this work bridges cutting-edge research with high-value, real-world use cases.

Open WebUI is a very popular component that provides a user interface to generative models. It supports use cases related to text generation, RAG, image generation, and many more. It also supports integration with remote execution servings compatible with standard APIs like OpenAI for chat completions and image generation. The goal of this project is to integrate Open WebUI with OpenVINO Model Server. It would include instructions for deploying the serving with a set of models and configuring Open WebUI to delegate generation to the serving endpoints.

This task involves various enhancements, such as: "Clone" Support for Edge Slide: An option for the Edge Slide tool which, when enabled, duplicates the corresponding edges first and then slides them. See RCS#2rcbbc, RCS#QDbbbc. Edge Flow: Adjusts the edge loop via spline interpolation such that it respects the flow of the surrounding geometry. See RCS#vddbbc. Loop Cut Curvature Preservation: Option added to loop cutting that when enabled places loop vertices on the curvature of the surrounding geometry with spline interpolation rather than with flat linear interpolation. See RCS#fSdbbc Expected outcomes: Additional loop-editing operators available in edit-mode. Skills required: Proficient in C++, Linear Algebra. Possible mentors: Campbell Barton, Tariq Sulley Expected project size: 350 hours Difficulty: hard

<p>In the systems biology field, there exists many different standards, with each standard serving for a different purpose. BioPax is a standard used for the representation of biological pathways. This standard has been traditionally used to describe structural and basic qualitative behavioral aspects of a biological design. While BioPax represents qualitative models, there are standards that describe quantitative models, namely the Systems Biology Markup Language (SBML). SBML is a standard for the representation of mathematical models of biological systems. SBML is built around core constructs that describe quantitative processes such as reactions. However, there are package extensions that allow for additional features. The Qualitative Models (qual) extension specifically allows SBML to describe qualitative models, which makes interconversion between both formats possible.</p>

Problem: Traditional anomaly detection systems often struggle with limited data scenarios and can lack the ability to explain their detections. Addressing these challenges is crucial for advancing the field and expanding its application range. Solution: This project proposes the integration of Vision Language Models (VLMs) into the anomalib framework using OpenAI's API. The goal is to enhance anomaly detection capabilities through few-shot and zero-shot learning techniques. This integration involves creating a new model class, 'LVM_openAI', as a subclass of 'AnomalyModule', and incorporating unique parameters like API_key and custom prompts. The model will leverage OpenAI’s API for processing and interpreting data, and will handle error catching, response parsing, and anomaly detection. Methodology: The project will start with a deep dive into Anomalib and OpenAI’s ChatGPT API, followed by a design and prototyping phase. The development will be in phases, with each phase focusing on different aspects of the integration, including testing with various datasets (e.g., MVTec AD and LOCO) and comparing against other models. Additionally, a comprehensive documentation, including a detailed notebook for setting up and using the model, will be created. Deliverables: An integrated VLM-based anomaly detection system within anomalib. Complete documentation and a set of experiments demonstrating the efficacy of the approach. An optional user interface for enhanced model interaction.

<p>Model Bricks project is an effort focused on describing recurring patterns in biological networks. These patterns are described as small VCell models that are retrieved from the VCell database and displayed online. To make Model Bricks recurring patterns more efficient and easily accessible, an entire code base shift from YAML to Vue.js is proposed. Vue.js features an incrementally adaptable architecture that focuses on declarative rendering and component composition. Vue.js component system will enable us to organize VCell APIs appropriately.</p> <p>The main objectives of this project are:</p> <ol> <li>Enhance Modelbricks.org user interface by using modern web design trends.</li> <li>Create a well-organized web architecture that will allow displaying of as many VCell elements as possible.</li> <li>Create VCell APIs for each element to be displayed.</li> <li>Enhance VCell user interface features to enable their display online.</li> <li>Write appropriate documentation.</li> </ol>

This proposal aims to create a production-ready, single-command deployment integrating the OpenHands AI coding agent with the OpenVINO Model Server (OVMS) on Intel AIPC hardware. The project resolves three critical friction layers: the API routing mismatch between OVMS (/v3) and OpenHands (/v1), the failure of structured tool-call execution which leads to model hallucinations, and system crashes caused by memory exhaustion from concurrent Docker sandboxes. By implementing a Docker Compose manifest with shared bridge networking and host Docker socket passthrough, the architecture offloads heavy LLM inference to iGPU or NPU targets while keeping orchestration on the CPU. Key technical deliverables include configuring native tool-guided generation for models like Qwen3-Coder-7B, enabling prefix caching to reduce multi-turn latency by over 50%, and enforcing strict container memory limits to ensure stability on 8 GB and 16 GB RAM profiles. The project concludes with a formal usability gap analysis and a finalized deployment setup to provide a seamless, local, and private AI development environment

The objective of this proposal is to design and implement an in-context AI assistant, named Co-Pilot, tailored for assisting users in creating Concerto models. This AI assistant will be seamlessly integrated into the existing VSCode web-extension, Playground, or other chatbot marketplaces, providing users with intuitive access to AI-driven suggestions and guidance during model creation. Co-Pilot will support multiple foundational AI models, such as OpenAI, Bard, Anthropic, among others, and allow users to configure API keys for their preferred models. Additionally, the proposal aims to engineer LLM prompts effectively to aid users in editing Concerto models and measure the effectiveness of these prompts.

<p>Systems biology researchers study the network of molecular interactions within a biological system and publish the deduced models in a diagram format. Many XML variants were developed to represent diagrams, including SBML and SBGN. However, SBML can represent much more complex models than SBGN. For example, the SBML qual extension package can mechanistically describe events in its models, such as the relationship that A activates C, B activates C, but A and B cannot activate C at the same time. The SBGN Activity Flow language can describe a similar relationship, but the relationship is represented visually only, and therefore is not stored in its model. The incompatibilities between SBML and SBGN prevent direct comparison between the models.</p> <p>In this project, I will build tools to support bidirectional conversion between SBML and SBGN, which acts as a machinery to map between SBGN-specific and SBML-specific information. SBML Layout package will be used to represent SBML diagrams, and the SBML qual package used to store the logic model.</p> <p>Blog: <a href="http://haleyyew-gsoc.blogspot.ca/" target="_blank">http://haleyyew-gsoc.blogspot.ca/</a></p> <p>Repository: <a href="https://github.com/NRNB-GSoC2017-SBML2SBGNML-Converters/SBML2SBGNML" target="_blank">https://github.com/NRNB-GSoC2017-SBML2SBGNML-Converters/SBML2SBGNML</a></p>

<p>Systems biology studies the mechanisms underlying complex biological processes as integrated systems of many interacting components. Modeling biological systems has always been a difficult and often challenging task. A particular challenge is to make such models reproducible and thereby reusable by a broader community. Different languages and libraries have been developed to address the exchangeable encoding of computational models. Examples are formats such as the Systems Biology Markup Language (SBML) and Simulation Experiment Description Markup Language (SED-ML), which can represent complex biological models and libraries such as the Systems Biology Simulation Core Library (SBSCL) which can simulate them. But until now, SBSCL still lacks certain features that need to be implemented. Essential new features are</p> <p>i) Implementation of new features of <strong>SBML Level 3 Version 2</strong>.</p> <p>ii) Support for newer versions of existing SBML packages such as <strong>FBC Version 3</strong> in SBSCL.</p> <p>iii) Implementation of <strong>stochastic</strong> simulation in SBSCL.</p> <p>iv) Improvement of support for SBML models with <strong>comp</strong> extension.</p>

<p>The <code>spatial</code> extension for SBML provides support for describing processes that involve a spatial component. This project seeks to implement validation functions for the <code>spatial</code> modelling package for SBML, thereby updating the extension to the latest specification. These functions will be used by the SBML offline validator to validate any models that contain a spatial component.</p>

This proposal outlines a detailed, structured, and actionable two-phase strategy for developing an OpenVINO AI PC Model Training Kit. The core objective of this initiative is to enable streamlined, efficient, and scalable AI model training capabilities on widely accessible hardware by fully leveraging OpenVINO’s optimization capabilities and integrating it seamlessly with leading machine learning frameworks such as PyTorch, TensorFlow, and Scikit-learn.

This project focuses on developing modular data preparation workflows within the PEcAn ecosystem to support ecological modeling.It targets the integration of diverse data sources for meteorology (e.g, NARR, ECMWF/ERA5, Ameriflux), soil (e.g, gSSURGO, SoilGrids, BADM), and vegetation initial conditions (e.g, LandTrendr AGB, BADM, bespoke CSV files)—and generate ensemble-based initial condition files. The goal is to create ensemble-ready inputs that enable sensitivity and uncertainty analyses. As a stretch goal, the project will implement an integrated workflow that runs the SIPNET model over these ensembles and performs variance partitioning to quantify the contribution of each data source to model output uncertainty. Current variance decomposition in PEcAn only supports parameter uncertainty. This project aims to address this gap by developing modular data preparation workflows that propagate and analyze uncertainty from multiple sources—including parameter, meteorological data, soil data and vegetation initial conditions. Deliverables include reusable R functions, data preparation modules, documentation, and example workflows integrated into the PEcAn system.

<p>This project aims at creating and implementing a system that automatically generates various spatial SBML models by comprehensively segmenting microscopic images of cells using deep learning and XitoSBML. The Systems Biology Markup Language (SBML) is a popular and widely accepted XML-based, open-source language for representing and exchanging models across analysis and simulation tools. By utilizing the capabilities of SBML via its spatial package, XitoSBML can create spatial SBML models from microscopic cellular images. However, this package has only been partially implemented and is still under development. A poignant requirement of this development process is a large number of test models, which can be provided as deliverables via this project to the SBML repository. Thus, this project will be able to contribute towards the development of the SBML Version 3 spatial package. Furthermore, the provision of several spatial SBML models will enable checking the compatibility of software tools that deal with spatial SBML models.</p>

The project aims to develop a robust pipeline for assessing the performance of foundation models, specifically focusing on scGPT, in generating embeddings suitable for inferring gene regulatory networks (GRNs) from single-cell RNA-sequencing (scRNA-seq) data. Leveraging the BEELINE benchmark and algorithms like GENIE3, GRNBoost2 and PIDC the pipeline will systematically evaluate scGPT's ability to capture gene regulatory relationships. Key steps include data preprocessing, embedding generation with scGPT, and GRN inference using GENIE3. Evaluation metrics such as precision, recall, F1-score, and area under the precision-recall curve (AUPR) will be employed, comparing inferred networks against simulated ground truth or expert-curated GRNs from BEELINE. The modular design of the pipeline allows for future integration of additional foundation models, enhancing its versatility and extendibility. This research will deepen our understanding of foundation models, particularly scGPT, in GRN inference, guiding the development of improved algorithms and methodologies for deciphering gene regulatory mechanisms from scRNA-seq data. Ultimately, this work will advance biological knowledge by shedding light on gene regulatory processes and their implications across various biological contexts.

In the realm of synthetic biology, there is a significant source of data in the form of models, genetic sequences, regulatory networks, and experimental results. Databases such as SynBioHub store this data represented by the Synthetic Biology Open Language (SBOL); using machine learning models, this data can be used by researchers to predict promoter activity, analyze protein-DNA interactions, and more. However, researchers must preprocess SBOL data before training their models, which can be time-consuming. Thus, this project aims to develop a Python package to automate the creation of datasets from data in SynBioHub. In doing so, we will first upload data from previous research to SynBioHub, create the package that returns data from SynBioHub in a format suitable for machine learning models, test the package on the data that was uploaded, explore the performance of a Graph Neural Network (GNN) on the data, and thoroughly document the package.

<p>In systems biology, it is very important to formalize the model for biological phenomenon and analyze based on this model basically constructed by ordinary differential equation(ODE). The model is normally described by SBML and there are several software tools to create the model easily. Regarding the analysis, this is considered as simulation, steady state analysis and parameter estimation. These methodologies are mathematically formalized and the library to analyze already exists. However, establishing the environment of analysis is difficult for biologists. Therefore, this is the obstacle to analyze ODE model easily. This project implement the web application to execute these analysis solving this problem. Web application does not require complicated establishment of analysis environment for users. This project removes the barrier for users to analyze ODE model.</p>

<p>This project aims at developing a tool which should provide an interactive visualization of the cancer progression models produced by the R library TRONCO. In particular, the web application will have two main functionalities: it will provide an interactive graph display and it will supply cancer progression models generation by directly calling the TRONCO functions. In addition to that it will provide the users with information about the genes involved in the model.</p>