Find open source projects actively accepting contributors. Search repositories, filter by program milestones, difficulty tags, or tech stack.
Use our Orbit AI Matcher to find out! Get instant matching scores based on your developer skills, preferred frameworks, and contribution experience.
Convert your selected open-source project into a winning GSoC, LFX, or Outreachy application using Proposal Studio.
This project aims to implement an in-memory, volatile file system for NVIDIA GPUs using the NVPTX backend in newlib. NVPTX offloading in GCC enables running tests on the GPU, but currently lacks proper file system support. The proposed solution involves two key objectives: (1) developing a lightweight file system driver that operates in the GPU’s VRAM, and (2) redirecting NVPTX system calls in newlib to interface with this in-memory file system. The file system will support basic operations such as file creation and deletion, without requiring recursive directory structures. This work will improve the runtime environment for NVPTX targets and potentially pave the way for future extensions, such as RPC-based host filesystem access or support for AMD GPUs via the AMD GCN backend.
This project aims to develop a hand-gesture computer control system by porting MediaPipe’s Gesture Recognizer to OpenVINO, enhancing performance on Intel hardware. It will have a context-aware system that maps hand movements to keyboard shortcuts, mouse clicks, and other inputs, with the innovation of application-specific mapping where identical gestures trigger different actions based on the active application. Deliverables include converted MediaPipe models (palm detection, landmark detection, gesture embedder, and classifier) in OpenVINO IR format with complete pipeline integration, a real-time gesture control system that achieves at least 20 FPS with smooth cursor tracking, and a user-friendly GUI for customizing gesture-to-input mappings; in addition, performance will be benchmarked across various Intel platforms and the implementation process will be thoroughly documented for future contributors.
This project aims to advance the evaluation framework for Google’s Gemini 2.0 and Gemma 3-27B multimodal models by integrating a diverse set of open-source, domain-specific benchmarks spanning healthcare, robotics, and general-purpose multimodal reasoning. In the healthcare domain, the evaluation will leverage datasets such as VQA-RAD and OmniMedVQA to assess the models' capabilities in interpreting and reasoning over complex clinical imaging and textual data. For the robotics domain, benchmarks like EmbodiedBench and EmbodiedQA will evaluate the models' proficiency in understanding spatial, visual, and language cues for grounded human-robot interaction tasks. Additionally, general multimodal benchmarks such as SME and BenchLMM will test the models’ ability to produce accurate, context-aware, and human-like explanations grounded in visual inputs. To ensure robustness, the project will also incorporate adversarial attack benchmarks across modalities, evaluating how resilient Gemini 2.0 and Gemma 3-27B are against input perturbations—including visual occlusions, textual prompt manipulations, and conflicting cross-modal signals. Alongside task accuracy, we will analyze the relationship between inference time and adversarial attack success, quantifying how model latency and confidence shift under adversarial conditions.
The clinical trial matching application aims to improve user-friendliness, particularly for individuals without medical expertise, by incorporating an algorithm to prioritize questions. This project focuses on optimizing and implementing the REACTJS web application component to render various question types with different input formats dynamically.
Overview This project aims to create a tool for Apache PMCs that will automate analysis of contributor activity in any given Apache repository. Project Goal The project goal is to make it easy for PMCs to manage their communities. We will try to achieve this by developing a tool that automatically finds and tracks first (or 3rd, or 10th) pull requests in a specified repository on Github that belongs to a specific community. The idea is that the PMC can take this data and send “Thank you” notes or emails to new people that joined the community and made their first contribution. If these automatic reports show that there are contributors with consistent commits to the project, PMC can send them encouragement emails to keep up the good work and become committers, etc. Project Description The tool will be based on using Github API to retrieve a list of all pull requests in the specified repository. Users will be able to set search parameters such as repository name and filters by pull request status. Users can use this tool to: Track pull requests in their own repositories Track issues opened in their own repositories Track the activity of individual contributors to encourage them to improve The tool will provide a user-friendly interface for searching and filtering valuable community information, such as pull requests, issues, questions in StackOverflow and others. Users will also be able to configure notifications to receive alerts about new pull requests. Technical Details The tool will be developed in Python and will use the Github API directly to fetch data (another possibility is using the PyGithub library to interact with Github API, however this may not be necessary). For the user interface, we can use the Flask or Django framework. The tool will be deployed on a web server and will be accessible through a browser.
Symmetry is one of the most beautiful and interesting phenomena in physics. Particle Physics is dominated by Lorentz symmetry. It is seen that Equivariant networks are more expressive and sample efficient than conventional Neural Networks. The project aims to build a Lorentz Equivariant Neural network to analyse particle collision data obtained from CMS Open Data. The project also aims to compare the results of the Lorentz Equivariant model with other Equivariant models(symmetry groups) and conventional CNNs and ViTs in terms of performance, speed and sample efficiency.
My primary focus for GSOC ‘22 at Zulip would be to enhance the UI/UX of the core Zulip Web Application. The web application UI is on its way to embrace modern design language while maintaining its solid performance and professional bug-free environment. I’m really excited to contribute to this initiative and put forward my efforts in making this proposal a success. I’ll be working on improving the UI/UX of the web application, implementing the redesign changes, making the CSS codebase maintainable and scalable, improving the Zulip settings and administrative pages, improving the accessibility, and working on improved markdown support. Besides this I’ll work on other high priority design, bug fixes and de-duplication issues, mainly in the front-end areas.
<p>3Dmol has most of the necessary features that make it a good visualization plugin for application and web. But this tool becomes difficult for the audience that is not familiar with web technology. Making its features inaccessible and making it harder to use as compared to similar applications like VMD, Visit, ChimeraX, PyMol etc. Hence there is a requirement of an UI that can make 3Dmol more accessible to the audience. To achieve this a layout for UI and additional controls for camera are proposed. That can expose all the features of 3Dmol. This will make 3Dmol easier to use and increase its application in general discussion, teaching and presentations.</p>
<p>Synthesis is at the heart of organic chemistry. It is the process of constructing a molecule through a series of reactions. Efficient molecular synthesis is significant, especially in areas like medicinal chemistry. However, it can be challenging to plan a synthesis for a complex molecule.</p> <p>Retrosynthesis is an iterative process of breaking down a target molecule into a series of simpler molecules which are readily available. It provides a systematic way to plan a chemical synthesis. But, as the space of possible chemical reactions is vast, it is hard for chemists to make the right disconnections while optimizing for multiple factors such as the cost of precursors and efficiency of the reaction.</p> <p>At its heart, the challenge of making a single retrosynthesis prediction can be recast as a Machine Translation task. Consequently, Natural Language Processing techniques can be leveraged to tackle this using Deep Learning.</p> <p>This project is focused on extending the DeepChem Library to support Retrosynthesis models, specifically to make single step predictions. This will be achieved by implementing a Molecular Transformer model and enhancing support for Reaction datasets.</p>
<p>This project proposes to refactor and rewrite most of the production code and tests to Kotlin, implementing all the modern features designed specifically to streamline mobile development, replacing the deprecated parts of the code, etc. resulting in safer and more maintainable code for future development.</p>
<p>The goal of this project is to add several recently developed spatial estimators to the exploratory spatial data analysis (esda) submodule of PySAL, the Python Spatial Analysis Library. This project will allow researchers to easily deploy these estimators in existing spatial workflows. Specifically, this project will contribute implementation, docstrings, tests, and example notebooks for (1) bivariate local join count statistics, (2) multivariate local join count statistics, and (3) local spatial heteroskedasticity (LOSH) statistics. In the first phase of this project, each estimator will be reviewed and pseudo-coded to identify areas of optimization. In the second phase, the estimators will be implemented with tests, with performance assessed against spatial datasets of different sizes. In the final phase, example notebooks will be drafted and polished to a quality ready for external (e.g. workshop) use.</p>
<p>The Oppia development workflow uses lint checks to help detect style errors before they reach the review phase. This project aims to refactor the linter and implement the remaining lint checks in the codebase and make the linting process more robust by adding tests.</p>
<p>Artists love to work with MyPaint brushes and it would be nice to integrate the same in Krita. The goal of the project is to integrate the MyPaint Brush Engine in Krita using the libmypaint library, so that Krita is able to load, save and work with MyPaint brushes. The secondary objective is to expose the MyPaint brush properties in Krita's brush editor to enable the creation and modification of MyPaint brushes.</p>
<p>Installer.to is a way of installing software on Linux, MacOS machines simply using cURL. Installer.to maintains a repository of publicly available bash script which installs software on different Linux distributions and MacOS. Anyone can add or update a script simply by sending a Pull request to GitHub.</p>
<p>Data retriever provides a number of publicly available datasets. These datasets are not stored at a single location but are downloaded from various publically available data repositories. New datasets are being added to retriever regularly so it becomes crucial to check the <strong>availability</strong> and <strong>installation</strong> of all datasets. This is a tedious task so the process needs to be automated.</p> <p>The goal of this project is to create a status server and dashboard where maintainers and users can see the status of datasets i.e. whether the datasets are installing properly or not and the changes that have been made to the dataset.</p>
<p>The AeroGear UnifiedPush Server sends native push messages to different mobile operating systems, such as Android, iOS and others. It has already developed good logical structures for sending of push notifications using JMS. Moreover, its "sender" module is mainly constructed of different message queues and their producers and consumers. On the other hand, the server DB is currently based on MySQL/PostgreSQL and each data entry lies in one DB scheme. We need the process for sending of push notifications and the evaluation of message metrics to be as fast as possible and as scalable as possible. Therefore, with an increase of our data problems can be caused by slow writes/reads to DB and not scalable enough message queue usage. The idea of the project is to try to overcome these obstacles and rebuild some modules of the system in the way that they perform well even if the data volume they are operating over increases. For that two different approaches can be taken. To speed up the DB access a new distributed, scalable, big data store HBase can be integrated into the project. For more scalable processing of the data, Apache Kafka can be used instead of JMS.</p>
<p>The InterMine Registry will allow users to obtain and search, by consuming a RESTful API, basic information about existing InterMine instances. This information includes instances names, URLs, versions, sizes, descriptions and principal data-types with their fields brief descriptions. Furthermore, the API is going to give the possibility to create and administer the registry entries.</p>
<p>Aim of the project is to implement the state of the art technique on Face Alignment i.e. "One Millisecond Face Alignment with an Ensemble of Regression Trees" by Vahid Kazemi using OpenCV version 3.2.</p>
<p>In this project I wish to develop a Mobile App for Mind The Word . I would be using Apache’s Cordova for developing the hybrid app as it would be easy to maintain and develop a cross platform app. Since the browser such as Chrome (for Android and iOS) do not support extensions, MTW cannot be implemented directly.In order to use our app, a user has 3 options.</p> <ul> <li>In the browser (e.g. chrome) , the user can choose to share the current page and then in the sharing options popup menu , the user can select MTW .</li> <li>The user can also copy the url to clipboard. Then as the user opens MTW App , a notification is shown at the bottom of the app with a message such as 'Open the copied url webpage ? ' . </li> <li>The user can manually enter the url in the address bar of MTW App.</li> </ul> <p>Any of the above 3 options, opens the InAppBrowser inside the MTW App and the mtw script is injected in webpage.The MTW App would include almost all the options (such as blacklist words, save learnt words etc ) available in the web extension.</p>
<p>According to the H.264/AVC standard, a decoder should be able to only decode the base view of an MVC encoded stream if the hardware doesn't support MVC decoding. The same applies to the base layer of a SVC encoded stream. I propose adding proper backwards compatible decoding support to vaapih264dec by resolving: <a href="https://bugzilla.gnome.org/show_bug.cgi?id=732267" target="_blank">#732267</a>, <a href="https://bugzilla.gnome.org/show_bug.cgi?id=732265" target="_blank">#732265</a>, <a href="https://bugzilla.gnome.org/show_bug.cgi?id=732266" target="_blank">#732266</a>.</p>
<p>The 5 advanced features added to the Chrome extension of the Wayback Machine are focused on users who wish to research the history of the web .</p> <ol> <li>The timestamp feature lets a user see the timestamp of external page elements in a playback page.</li> <li>The radial tree feature lets a user visualize the hierarchy of pages in a website as a radial tree also known as a sequences sunburst diagram.</li> <li>The 'get books' feature lets a user navigate to a book on Internet Archive by analyzing a Wikipedia page for book titles. </li> <li>The ORCID feature displays the ORCID accounts of people whose names match a name given as input.</li> <li>The robust links feature is useful for web developers who wish to fight the phenomenon of link rot. This feature extends the Memento Project by Los Alamos National Laboratory and Old Dominion University.</li> </ol>
<p>My project would re-implement the <a href="https://github.com/bundler/bundler-api" target="_blank">Bundler API</a> on the <a href="https://github.com/rubygems/rubygems.org" target="_blank">Rails app that provides RubyGems.org</a>. Bundler uses bundler-api to download dependency information about all the gems it is trying to install. This request for complex database queries, marshalling the output and everything else is quite CPU intensive and rubygem.org limited hardware infrastructure could not support it until now. With help of <a href="https://rubytogether.org/" target="_blank">Ruby Together</a>, hopefully we will make <code>bundle install</code> faster and more reliable for everyone.</p>
This project addresses the challenge of unreliable LLM outputs in complex health informatics workflows by building a domain-grounded Retrieval-Augmented Generation (RAG) system within JuliaHealth. It enables users to convert natural language queries into accurate, executable FunSQL.jl workflows by grounding responses in curated resources like OMOP and OHDSI. The project will deliver a modular baseline RAG pipeline, a reproducible evaluation framework to benchmark query quality, and clear documentation with an end-to-end demonstration for usability and future extension.