Explore Projects

This project focuses on implementing an enhanced Model Context Protocol (MCP) and A2A communication framework for KubeStellar's Management Control Plane server. It builds upon the foundation MCP implementation to include advanced A2A coordination capabilities, distributed AI agent communication, and sophisticated context management for multi-cluster environments. Expected Outcome: A fully specified Model Context Protocol with A2A communication extensions, implementation of protocol handlers in Python for the MCP server, A2A communication framework enabling AI agent coordination, context management system for KubeStellar state information, serialization mechanisms for cluster state, protocol extension mechanisms for different AI providers, performance optimizations, and comprehensive test suite.

In Zulip, each message has a sender (linked to UserProfile) and a recipient (referencing the Recipient model). The Recipient model supports three types: 1. Personal -- linked to a single UserProfile 2. Stream -- linked to a Stream 3. Direct Message Group -- linked to a DirectMessageGroup Currently, 1:1 direct messages use the Personal type that directly links recipients to individual users. This project aims to migrate them to use the Direct Message Group type instead. The migration should be seamless with minimal downtime. It involves: 1. Updating the codebase to support 1:1 DMs using DirectMessageGroup. 2. Running a backfill migration to create a DirectMessageGroup for each user pair with existing 1:1 messages. 3. Cleaning up legacy code and adding additional fields that are needed. This migration enables per-conversation settings for direct messages, and features like pinning or hiding DM conversations can be implemented more easily.

<p>Recently various nonlocal models have been applied for understanding of complex spatially multiscale phenomena like solid mechanics, fluid mechanics, particulate media, directed self-assembly of Block-Copolymer, etc. Nonlocal models have dependence of a single point on a small set of near neighboring points, which hints for data-level parallelism. We utilize a modern parallelization library, HPX, to develop the parallel solver. The goal of the proposed solution is to achieve full distribution of workload starting from mesh partition to computation of fields at mesh nodes. The following goal is achieved step by step, starting from a simple sequential implementation of 1D nonlocal diffusion. The 1D nonlocal diffusion equation is made multi-threaded to enable concurrent processing within a single node and then is extended to a 2D nonlocal diffusion equation. The 2D nonlocal diffusion equation is then made completely distributed to exploit data level parallelism.</p>

Researchers have tremendous difficulty keeping up with the exponential growth of scientific literature--where groundbreaking findings and complex concepts are emerging at a rapid pace--hindering their ability to stay on top of the latest insights in their field. There is a dire need for AI utilizing retrieval-augmented generation (RAG) models that can efficiently process and summarize these vast volumes of scientific papers, helping human researchers focus their efforts on the most promising areas of investigation. The "AI Scholar" project aims to develop an advanced AI system utilizing state-of-the-art natural language processing, information retrieval, and machine learning techniques, including a RAG model, to efficiently process, summarize, and analyze vast volumes of academic papers. This innovative system will enable personalized recommendations, automated literature analysis, idea generation, and guidance for researchers navigating the complex landscape of statistics, data science, and potentially other scientific domains, thereby accelerating the pace of scientific discovery and innovation.

Toolkit for Multivariate Analysis (TMVA) is a multi-purpose machine learning toolkit integrated into the ROOT scientific software framework, used in many particle physics data analysis and applications. Currently, we are developing a fast inference system in TMVA, called SOFIE, that takes takes ONNX model as input and produces compilation-ready standalone C++ scripts as output. These scripts will then provide users an easy way to deploy their deep learning models in their physics software and analysis frameworks. This project will focus on development of some missing deep learning operations which will allow to build more complex networks within TMVA. Specifically, we propose to implement the inference functionality of some ONNX operators in the code generation format. The student can choose to build this based on existing implementations in TMVA or other existing machine learning software tools or build their own from scratch. The expected result is a working implementation of modular operators classes that implement the operators as defined by the ONNX standards in the code generation format. The project requires also to write the corresponding unit tests need to validate the written code

Embed Wasm runtime filters into Vector data routing topologies for custom metric transformations at line rate.

<p>Extending Neural SPARQL Machines (NSpM) to cover more DBpedia classes to enable high quality Question Answering</p>

Last year there was some work regarding Avogadro's usability as a tool for bioinformatics. My proposal is working along the same line, giving Avogadro the ability to display molecular surfaces, non-covalent bonds, and handle other molecular features like coordination, all while improving the existing code's speed and quality.

Ganga routinely handles a tremendous amount of computational tasks simultaneously and keeping track of their status is prone to lagging behind with its current implementation, particularly regarding remote jobs submitted through DIRAC. Furthermore, it is the project mentors’ belief that the current implementation of the monitoring service could be simplified and brought up to date. My goal at the end of the project’s timeline would be to deliver a new and efficient monitoring service using the asyncio framework. The main goal is for each backend’s monitoring to be rewritten using asyncio to make it as efficient as possible, but the service will also support multithreading and multiprocessing executors. It will be backend agnostic and allow any new backend to easily hook up to it. I also plan to specifically address DIRAC’s current inefficient monitoring implementation by using its REST API to retrieve running job information.

<p>The main objective of this project is to enhance the flexibility and versatility of Sigmah software to meet new requirements and satisfy user feedback and will be working in this regard as the GSoC project. The Sigmah's concept is extremely fascinating and can be worked over so that it accomplishes greater heights.</p>

EDM4hep offers a powerful set of objects to describe event data from simulation to analysis. At any step of this chain, one can inspect the event data containing various collections of the objects, e.g. calorimeter clusters, tracker hits, reconstructed particles, etc. At the moment dmX offers only a limited possibility to visualize a tree of MonteCarlo particles. In this project, dmX is extended to visualize other types of collections and relationships between them as a graph to gain more insights about the collisions useful for further development in the Accelerators world.

<p>ZAP allows the penetration tester to set up authentication for the web application being tested. This allows ZAP to run tests from the point of an authenticated user. This is an excellent feature as with the ability of session management and automatic (re)authentication ZAP can check for vulnerabilities that are hidden in authorized resources. This is also a good feature to test broken authentication and session management.</p> <p>However, as of now setting up authentication in ZAP is a multi-step process [1]. It is time-consuming and error-prone. This project targets to resolve this problem by adding three new features to ZAP.</p> <ol> <li>Automated configuration</li> <li>Guided configuration</li> <li>Authentication status scanning</li> </ol> <p>[1] <a href="https://github.com/zaproxy/zaproxy/wiki/FAQformauth" target="_blank">https://github.com/zaproxy/zaproxy/wiki/FAQformauth</a></p>

<p>Project aims at implementation and proof of concurrent on-the-fly SCC detection algorithm to aid in liveness checking of TLC model checker. The goal is to allow TLC's liveness checking to scale on large state space as opposed to Tarjan's sequential solution to this problem.</p>

This project is based on the proposed project (Learning Loop Transformation Policies) by the mentors Johannes Doerfert, Mircea Trofin. This project would like to experiment on how the compiler may learn for improvements by observing from the back-end. We can try it on RISC-V, an open instruction set architecture.

<p>Symbiflow has already produced a model of Xilinx's Artix7 in VTR, but it only limited to single chip description. The aim of this project is to instead investigate how to describe Artix7 at a higher level of abstraction, using VTR's architecture description language, so the chip definition can easily be changed to make conducting experiments with related architecture easier.</p>

CVXPY is a Python-embedded modeling language for convex optimization problems. In this project, I aim to add support for new advancements in optimization methods for problems involving the ubiquitous matrix logarithm --- which in this context has especially important applications to quantum information. These new tools will dramatically expand CVXPY’s usefulness to QI theorists and quantum computer engineers.

Large Language Model (LLM) serving frameworks like vLLM must deliver high throughput and low latency to justify the substantial cost of GPU and TPU deployments. This project targets systematic benchmarking, profiling, and optimization of vLLM, one of the most widely adopted open-source LLM serving systems. The goal is to push its performance limits under realistic workloads and contribute meaningful improvements back to the community.

<p>LanguageTool (LT) currently uses neural networks to detect confusion between words. So far, it only considers 2 words of context in both directions (tri-grams). The aim of this project is to extend this to take the entire sentence as context to detect the confusion pairs. For this, the project aims to leverage the use of a sequence-to-sequence model to detect the word(s) that are used incorrectly and then replace them with the correct word(s). The project aims to give this functionality to English, French, and German.</p>

Ganga is a tool for composing, running, and tracking computing jobs across a variety of backends and application types. Ganga primarily runs as a command-line tool and has an IPython-like prompt. My proposal lays out a plan to add a large language model to the Ganga prompt. It discusses options like cloud serving, local serving, and the challenges that come with each. It also addresses whether a rag system would make sense for the project at this juncture. Finally, I suggest a timeline breakup where I first focus on implementing a local solution followed by a cloud one.

This project proposes integrating ACTINN, a neural network for automated cell type identification in single-cell RNA sequencing (scRNA-seq) data, into the DeepChem framework. ​Integrating ACTINN into DeepChem would mark the framework's inaugural inclusion of a single-cell deep learning model, significantly expanding its capabilities in the burgeoning field of single-cell analysis. By incorporating ACTINN, DeepChem can offer researchers a powerful tool for automated cell type identification, streamlining workflows in single-cell analysis and fostering deeper insights into cellular heterogeneity.

Perception metrics is a unified toolkit for evaluating perception models across datasets and frameworks. Its foundation was built through prior GSoC efforts designed to unify and formalized in a stable release published in Sensors. This platform was recently modernized to support LiDAR, images segmentation, and object detection. This proposal focuses on scaling PerceptionMetrics from off-road focused workflows to industry-standard benchmarking. The core objectives are: • Integrate and prioritize standard image and Lidar datasets. • Extend GUI support to segmentation visualization (image and Lidar). • Deliver comprehensive tutorials/documentations for practical usage. • Strengthen reliability through expanded tests and CI-friendly validation.

<p><strong>Develop tools for performance evaluations of machine learning models</strong></p><p>The deadline to create an initial application has passed for the May 2020 Outreachy internship cohort. We are no longer accepting initial applications for internships. We encourage you to sign up for the announcements mailing list to get an email when the next round opens. Why apply to Outreachy? &gt; Start your initial application &gt; May 2020 Outreachy internship cohort &gt; Mozilla Community details are hidden until you are approved to participate as a mentor or coordinator.</p><p><br></p><p><strong>Mentorship Cohort:</strong> 2020</p>

<p><strong>Improve the documentation and outreach material of the Ersilia Model Hub</strong></p><p>The deadline to create an initial application has passed for the May 2022 Outreachy internship cohort. We are no longer accepting initial applications for internships. We encourage you to sign up for the announcements mailing list to get an email when the next round opens. Why apply to Outreachy? &gt; Start your initial application &gt; May 2022 Outreachy internship cohort &gt; Ersilia Open Source Initiative Community details are hidden until you are approved to participate as a mentor or coordinator.</p><p><br></p><p><strong>Mentorship Cohort:</strong> 2022</p>