Contribo
ProgramsOrganizationsProjectsProposal StudioResourcesOrbit AIDashboard
Login
Login
Contribov1.0

Your premium companion for discovering, matching, and tracking paid open-source mentorship programs and verified starter issues.

Product

  • Programs Directory
  • Orbit AI
  • User Dashboard

Resources

  • Guides & Resources
  • Proposal Studio

Community

  • GitHub Repository
  • Code of Conduct

Legal

  • Privacy Policy
  • Terms & Conditions
  • Cookie Policy
  • Accessibility Statement

Made with love bySubhranshu Sekhar Dash

© 2026 Contribo. Built for the global open-source community.

PlatformProjects

Explore Projects

Find open source projects actively accepting contributors. Search repositories, filter by program milestones, difficulty tags, or tech stack.

Confused which project to choose or don't know which one you are most suitable to work on?

Use our Orbit AI Matcher to find out! Get instant matching scores based on your developer skills, preferred frameworks, and contribution experience.

Match with Orbit AI
Filters Clear
3179 Projects Found
Sort by

End to End Open Verifiable LLM: Deterministic Training & Verification

intermediate
AOSSIE•2026Varies

Ready to Apply?

Convert your selected open-source project into a winning GSoC, LFX, or Outreachy application using Proposal Studio.

Start Proposal
Proposal Status
0% ProgressNot Started
Accepted Proposals
3 SimilarView →
Proposal Guide
Tailored DocsOpen →
Checklist & Review
Readiness AuditCheck →
175h (Medium)

Open-weight LLMs are reproducible in principle but not verifiable in practice. You can download weights, but you cannot prove what data they were trained on or whether they were modified later. This project closes that gap by building a training pipeline where identical configurations produce identical model states, and any deviation is cryptographically detectable. A key insight from my experiments is that while training computation can be deterministic, PyTorch’s .pt serialization breaks verification: identical weights can produce different checkpoint files. This project resolves that by verifying tensor-level state using byte-stable formats rather than file-level artifacts. The deliverable is a 1B–3B model trained on a pinned Wikipedia dataset, along with a public manifest that enables independent reproduction and audit.

pythonjavascriptswift+8
Student: ryoari
Code
Proposal

STOFS Subsetting Tool

advanced
IOOS•2024Varies350h (Large)

Developing a subsetting tool to access ocean forecast data is critical and addresses an immediate need for maximizing the utilization of STOFS model output. STOFS-2D global stands out as one of the most refined and high-resolution global ocean forecast datasets available. The STOFS Data Subsetting Tool project enhances the xarray-subset-grid library to efficiently subset STOFS and oceanographic data. Through Python libraries like Xarray, Zarr and Dask, it benchmarks performance, integrates with ocean modeling packages, and develops a service for on-demand subsetting via email or API. The project spans 12-16 weeks, focusing on method benchmarking, performance evaluation, optimization, service development, and the creation of a portal for subsetting datasets.

pythonocean sciencejava+3
Student: Omkar Kabde
Code
Proposal

Alternatives to Neighborhood-Based Collaborative Filtering

advanced
mlpack•2018Varies350h (Large)

<p>The overall objective of this project is to improve the current CF module in mlpack to provide better rating prediciton, fast execution, and flexible APIs.</p> <p>To be more specific, the objectives include:</p> <ol> <li>Solve the current problems which are affecting the accuracy (e.g. lack of data normalization, lack of alternative methods for aggregation of neighbor ratings).</li> <li>Add CF models which are more expressive, and models which can take implicit feedbacks into consideration (e.g. BiasSVD, SVD++).</li> <li>Benchmark the cf module with public datasets (e.g. MovieLens, Netflix).</li> <li>Accelerate program execution by identifying and modifying the codes which are slowing down the execution.</li> </ol>

c++pythonmachine learning+4
Student: Wenhao Huang
Code
Proposal

PyMC - Fast Exact Gaussian Processes

advanced
NumFOCUS•2022$6k–$13.2k350h (Large)

Gaussian Processes (GPs) are a flexible class of probabilistic models which can automatically fit quite complex functions. In addition, GPs provide a principled means of computing the predictive uncertainty. PyMC includes an implementation of GPs as a result of a previous GSoC contributor. However, Gaussian processes are notorious for being very expensive to compute. Training a GP scales cubically with the training data points. This gets prohibitively expensive even for a few thousand training data points. There has been quite a lot of recent research in speeding up the training of GPs and this proposal will implement plans to implement a recently proposed GPU accelerated exact GP model [1](https://proceedings.neurips.cc/paper/2018/hash/27e8e17134dd7083b050476733207ea1-Abstract.html)

pythonjavascriptr+3
Student: KunalGhosh
Code
Proposal

Rewrite for Turn Restricted Shortest Path Algorithm in PgRouting

intermediate
OSGeo•2017Varies175h (Medium)

<p>In graph theory, the shortest path problem is the problem of finding a path between two vertices (or nodes) in a graph such that the sum of the weights of its constituent edges is minimized. In real life scenario, the road network is modeled as a graph where the graph's vertices correspond to road junctions and the edges correspond to road segments, each weighted by the positive length of its road segment.</p> <p>A turn models a movement from one edge element to another. Often in the real world scenario, turn restrictions such as no-left-turn, no-right-turn etc. are imposed on the road network.</p> <p>The aim of the project was to come up with a solution for calculating the shortest path from a given source node to a destination node in the graph containing turn restrictions.</p>

cpythonjava+6
Student: Vidhan Jain
Code
Proposal

VLC Qt interface redesign

intermediate
VideoLAN•2019Varies175h (Medium)

<p><strong>To rework the VLC interface heavily and make it beautiful and useful again.</strong></p> <p>I'll be working on redesigning the new interface(VLC 4.0) to make it feature compliant with the old interface.</p> <p>This includes making Qt models to transfer data from/to the different VLC components. The data in the models will be accessible via a UI which will be created using qml.</p> <p>More specifically i'll be working on:</p> <ul> <li>The video view tab which has:<ul> <li>Videos</li> <li>Shows</li> <li>Camera Roll</li> </ul> </li> <li>Internet tab that hooks up with:<ul> <li>Radios </li> <li>Podcasts</li> </ul> </li> </ul>

copenglc+++10
Student: Abel Tesfaye
Code
Proposal

QR/Barcode/ArUco detector

beginner
OpenCV•2025Varies90h (Small)

Description: QR, Barcode and ArUco are all popular code in computer vision applications. OpenCV now support all of them, and can detect and decode them. But OpenCV still expect a better detector and decoder for them. If possible, one efficient deep detector for all kinds of codes can simplify the usage. If one efficent deep detector cannot be achieved, several deep models are also acceptable. Expected Outcomes: Train a deep detector for QR, Barcode and ArUco. Or train three different deep detectors for different codes specifically. The trained model should be easy to implement with the current algorithms in OpenCV on QR/Barcode/ArUco. A nice demo to show how to use the algorithm. Detailed report to demontrate if the trained detector(s) are better than the current solution in OpenCV.

pythonmachine learningc+++9
Student: Jinyue CHEN
Code
Proposal

Enhanced AI OCR Extraction Pipeline for Scientific Literature

advanced
Open Science Labs•2025Varies350h (Large)

This project creates a comprehensive pipeline for extracting text and complex tables from scientific literature PDFs. By combining PyMuPDF, specialized table detection models, and Vision-Language Models (VLMs), the system will significantly improve accuracy over traditional OCR methods while maintaining cost efficiency. The pipeline allows users to upload multiple research papers, select their preferred extraction method, and receive structured markdown outputs with preserved document hierarchy. A side-by-side interface will enable human verification and correction of extraction results. The solution will be optimized for lightweight deployment on platforms like Hugging Face Spaces, making it accessible to researchers worldwide while focusing primarily on complex table formats common in life science research.

pythonjavascriptc+++3
Student: Priyankesh
Code
Proposal

ML Inference on heterogeneous architectures using SOFIE

intermediate
CERN-HSF•2026Varies175h (Medium)

SOFIE generates standalone C++ inference code from trained neural networks, letting HEP frameworks run ML models without heavy external dependencies. Its integration with the alpaka library enables portable GPU inference that works across NVIDIA, AMD and CPU backends from a single codebase. The current GPU path covers basic operations but lacks support for several key operators needed to run full neural networks on GPU. This project implements the missing GPU kernels and extends convolution to handle batched inputs, so complete CNN architectures can run end-to-end on GPU through SOFIE. The final goal is a ResNet-style model running fully on the alpaka GPU backend and validated against the CPU reference.

pythonc/c++machine learning+6
Student: _HarshChauhan_
Code
Proposal

Machine Translation Metric leveraging FrameNet

intermediate
FrameNet Brasil at the Federal University of Juiz de Fora•2019Varies175h (Medium)

<p>Machine Translation is one of the most essential tasks of Natural Language Processing where a lot of research has been done. There are various metrics that are available to evaluate the quality of translation but most are obtained by computing the similarity between an MT hypothesis and a reference translation based on character N-grams or word N-grams. Using the data generated from the annotation of the TED Talk in two or more languages, as well as the relations between frames given in the Berkeley FrameNet Data Release 1.7, our task for this project is to build an automated metric system for machine translation, which is intended to measure the frame distance between sentence pairs in two languages. In this proposal, the task has been modelled as a regression task of predicting accuracy scores. The frames evoked and the related frame network has been used as pivotal features of the model.</p>

pythonmysqljavascript+1
Student: Debanjana Kar
Code

Robotics-Academy: Improve Deep Learning based Human Detection exercise

advanced
JdeRobot•2022$6k–$13.2k350h (Large)

This project aims to improve the “Human detection” Deep Learning exercise at Robotics-Academy, developed along GSoC-2021. Instead of asking the user to code the solution in a web-based editor, they will have to upload a deep learning model that matches video inputs and outputs the image coordinates of the detected humans. In this project, the support for neural network models in the open ONNX format, which is becoming a standard, is one goal. The fluent exercise execution is also a goal, hopefully taking advantage of GPU at the user’s machine from the RoboticsAcademy docker container. An automatic evaluator to test the network's performance provided by the user will also be explored. All the above functionalities have, to some extent, already been implemented in the exercise. The goal is to improve/refine them either on the existing implementation or from scratch.

pythonrosgazebo+2
Student: Zekrom_7780
Code

Generating Hessians and Jacobians via CLAD

intermediate
CERN-HSF•2019Varies175h (Medium)

<p>Clad is a C++ Clang compiler plugin that employs automatic differentiation to derive user-defined functions, performing source code transformations so that users do not have to conform to custom types to comply with external libraries.</p> <p>With the integration of clad::gradient to CLAD, a reverse accumulation method for automatic differentiation was introduced. Now, it makes sense to move on to second partial derivatives, in particular, to calculating the Hessian matrix. My work is to build on the existing framework that uses Clang AST to do source transformations on functions, and to implement an efficient Hessian calculation method that extends the capabilities of CLAD using the edge pushing algorithm and Hessian reverse accumulation method. I will also try to extend existing CLAD functions to calculate the Jacobian matrix.</p>

pythonc/c++machine learning+6
Student: Jack Qiu
Code
Proposal

Reactive Loan Risk Assessment Engine for Mifos

intermediate
The Mifos Initiative•2025Varies175h (Medium)

Provide financial institutions with real-time insights into loan risk, helping them make better lending decisions. A reactive backend ensures the system can handle a growing number of requests as Mifos deployments expand. Improved risk assessment can lead to more responsible lending practices, ultimately benefiting underserved communities. Create a dedicated microservice in Java that handles loan risk assessment asynchronously. Utilize reactive programming paradigms to ensure scalability and high throughput. Design and implement a configurable risk scoring model that can incorporate multiple factors such as client financial history, repayment behavior, and external credit ratings. Allow for model tuning via configuration, so institutions can adapt the scoring criteria to local requirements.

mysqljavascriptandroid+5
Student: Abhinav R Cillanki
Code
Proposal

Adding New Plots to ArviZ

intermediate
NumFOCUS•2021Varies175h (Medium)

<p>ArviZ is a python package for exploratory analysis of Bayesian models. It is designed as a backend agnostic tool and supports different backends for visualisation and diagnostics. Apart from plotting ArviZ also supports functions for posterior analysis, model checking, comparison and diagnostics.</p> <p>As ArviZ is mainly a visualisation library, it already has various plotting functions to analyse Bayesian Inference data. Still, it would be nice to extend its visualisation capabilities and add some more plotting functions which the community could use.</p> <p>This project aims to implement dot plots, quantile dot plots, half-eye plots, ecdf plots, and, if time permits, then also calibration plots for classification into ArviZ for both the backends Matplotlib and Bokeh and also to write tests for all these plot functions.</p>

pythonjavascriptr+3
Student: Rishabh Sanjay
Code
Proposal

VectorFlow

advanced
CERN-HSF•2019Varies350h (Large)

<p>A C++ adapter API for integrating vectorized components in a scalar workflow.</p> <p>Many FLOP-intensive algorithms may profit from the vector pipelines of modern processors; they don’t because they don’t have vectorizable inner loops. The project idea is to implement and benchmark a generic vector flow service that can non-intrusively integrate with arbitrary data processing frameworks and can expose algorithms to the higher-level event loop of these frameworks.</p> <p>This library aims to provide an API and a set of examples demonstrating the transformation of the scalar algorithm in a vectorized one. Depending on the intrinsic algorithm gain from SIMD vectorization and better data caching, the overheads introduced by the extra data transformations can be much smaller than the benefits.</p>

pythonc/c++machine learning+6
Student: Arturo Garza Rodriguez
Code
Proposal

Relational Framework for Prompt-based Automatic Layout Generation with Glayout

beginner
CHIPS Alliance•2024Varies90h (Small)

The proposal seeks to build upon an existing relational framework that facilitates the generation of DRC-compliant analog layouts through conversational English prompts. It addresses challenges in analog layout creation, such as the necessity of coding expertise with Glayout API or the time-intensive nature of GUI-based approaches, as well as the need for automatic layout generation from high-level circuit descriptions. The strategy involves rectifying existing framework bugs and enhancing layout validation through Magic DRC and LVS checks. Subsequently, the framework will be reinforced by simplifying routing and enabling streamlined creation of diverse circuits in a single step. Finally, the framework will be expanded by integrating in a Large Language Model (LLM) to translate very general conversational prompts into simpler commands usable by the relational model.

fpgachiselrisc-v+2
Student: Chetanya Goyal
Code

ABI/Layout handling for the automatic differentiation feature

advanced
The Rust Foundation•2025Varies350h (Large)

The integration of automatic differentiation (autodiff) into the Rust compiler via Enzyme faces challenges due to ABI adjustments that alter function signatures. These transformations, while beneficial for performance, interfere with autodiff functionality, leading to incorrect differentiation. This project aims to prevent unwanted ABI adjustments for functions marked with #[rustc_autodiff], ensuring autodiff functions maintain their intended behavior without affecting other optimizations. The approach involves modifying Rust’s function ABI handling to recognize and preserve differentiation-related information. Key steps include identifying problematic transformations, encoding autodiff metadata in FnAbi, and preventing ABI changes that break autodiff. By addressing these challenges, the project will improve autodiff support in Rust, making it more reliable and consistent while preserving compiler efficiency.

pythonrust
Student: Sa4dUs
Code
Proposal

Project Proposal for "Graphical interface for inserting plugins during simulation" Causevic

beginner
Open Source Robotics Foundation•2016Varies90h (Small)

<p>In order to add a new plugin Gazebo programmers must change the .world file and restart Gazebo each time [15]. Furthermore modifying source code, changing plugin parameters or attaching plugins to new models has the same effect. Since this is a rather tedious and time-consuming process and users are thereby distracted, they can not fulfill all tasks in one run. A new idea favoring an easy-to-use GUI and a simple plugin implementation concept should eliminate this procedure. Coupled with a number of additional innovative features this could lead to a behaviour that end-users easily view attached plugins on specific models, interact lightning-fast with various plugin types, e.g. GUI, visual and world plugins or modify plugin source code at runtime.</p>

rosgazebopython+3
Student: AlminCausevic
Code

AI-Assisted Curation Assistant Tool for cBioPortal

beginner
cBioPortal for Cancer Genomics•2026Varies90h (Small)

This project aims to extend the existing cBioAbstractor prototype into a schema-driven formatting pipeline that assists curators in transforming heterogeneous supplemental datasets into cBioPortal-compatible formats. Currently, curation is a manual and time-consuming process due to inconsistent file structures, lack of automated schema mapping, and the need to interpret validation errors manually. This results in inefficiencies, variability, and high onboarding effort. My preliminary extension implementation suggests the feasibility of this approach by connecting existing components, introducing fuzzy column normalization, and providing a simple interface for file upload, transformation, and result inspection. Initial qualitative experiments show that the system can reconstruct correct cBioPortal formats from intentionally perturbed datasets, suggesting strong potential for reducing manual work and improving consistency.

javascriptjavaweb+9
Student: fppcng
Code

Complex Embeddings for OOV Entities

advanced
DBpedia•2018Varies350h (Large)

<p>The aim of this project is to enhance the DBpedia Knowledge Base by enabling the model to learn from the corpus and generate embeddings for different entities, such as classes, instances and properties. While we do this, it is imperative that these embeddings are able to accommodate the semantic relatedness between entities. This means that we are not limiting ourselves with just the similarity between words, instead we take a step further ahead to also define the relatedness between the vectors and thus the relation between the entities and the text. Therefore, to incorporate this measure of the semantic distance, we define a measure of descriptiveness of the class that these entities belong to. Entities belonging to a class that has a very high level of description must have very low semantic distance in our model. Eventually, we extend the usability by predicting embeddings for out-of-vocabulary entities as well, and also extract relations between those entities using approaches that have been previously used for link prediction tasks in machine learning.</p>

javascalardf+7
Student: Bharat Suri
Code
Proposal

GlitchWitcher & VitAI: AI-assisted Triage and Bug Prediction

advanced
Eclipse Foundation•2025Varies350h (Large)

This project successfully developed two AI-powered tools to enhance software quality at Eclipse Adoptium. The first, GlitchWitcher, implements the REPD (Reconstruction Error Probability Distribution) model, using an autoencoder for supervised anomaly detection to proactively predict software defects. The project's primary deliverable evolved into VitAI, an advanced, modular multi-agent AI assistant for interactive test triage. VitAI's architecture features a custom MCP (Model Context Protocol) server for providing an interface to interact with a specialized agent and a Retrieval-Augmented Generation (RAG) system. The RAG system provides deep, context-aware responses by querying a vector database of project documentation, issues, and build data. This work was supported by automated data pipelines and is showcased in a live interactive demo.

c/c++javageospatial+15
Student: Anirudh Sengar
Code

EchoGem – Teaching Gemini to Think in Batches by Prioritizing What Matters

intermediate
Google DeepMind•2025Varies175h (Medium)

EchoGem introduces a novel batching engine designed to answer multiple questions about the same source parallelly to reduce response times heavily. It does so by building a modular smart batching engine for Gemini that teaches the model to think in context-aware batches. What sets EchoGem apart is its focus on modularity and testing—each component is designed to be independently swappable and improvable allowing for open testing and matching different strategies for different parts of the engine. Instead of a naive sliding window or top-k search approach, EchoGem uses semantic clustering of both chunks and questions to form intelligent batch groups. It is also committed to measurable, reproducible gains. Every design decision—whether it's a new chunking strategy or a different context ranking model—can be rigorously backtested using a structured evaluation suite.

pythonjavascripttypescript+2
Student: Aryan Saboo
Code

Interfaces for consistent statistical analysis in the Python Spatial Analysis Library (PySAL)

intermediate
NumFOCUS•2022Varies175h (Medium)

For many users of statistical software, the most important aspect of a statistical software package is convenience. Therefore, it is important to structure statistical software libraries in a natural way that reflects how users most commonly think about their problems. At present, many of the user-facing interfaces of the Python Spatial Analysis Library (PySAL) follow inconsistent design patterns and may not be the most natural structures in terms of ease of use. This 350-hour project will study consider a set of interface designs for standardized use across the library’s statistical modeling packages and implement one such design in a development version of the library. If time permits, another aspect of this project that may be tackled is the extension of R-style Wilkinson formulas to spatial models via a Python package such as patsy or formulaic (Wilkinson and Rogers, 1973).

pythonjavascriptr+3
Student: Tyler Hoffman
Code

Async Support for TensorFlow Backend in FFmpeg

advanced
Intel Video and Audio for Linux•2021Varies350h (Large)

<p>This project focuses on implementing an asynchronous mechanism for model inference and batch execution in the TensorFlow backend of the FFmpeg Deep Neural Network module to boost model inference performance.</p> <p>The Tensorflow backend uses the TensorFlow C API, which currently does not provide functions for asynchronous execution. The support for async behavior can be provided using multithreading on the existing TensorFlow library functions. We will implement this behavior through detached threads that work independently of each other.</p> <p>Several inference frames will be combined to a single input tensor and executed together in a single batch to enable the batch mode. The DNN module authors saw a performance gain in the OpenVino backend with asynchronous batch inference against synchronous inference. A similar performance gain is expected from this project.</p>

gstreamerffmpeglibxcam+6
Student: Shubhanshu Saxena
Code
Showing 2065 - 2088 of 3179 Projects
8586878889
Proposal
Proposal
Proposal
Proposal
Proposal
Proposal
Proposal
Proposal
Proposal