Back to Programs

Infrastructure and cloud

Kubeflow

The Machine Learning Toolkit for Kubernetes (AI)

YAMLdevopsgojavascriptkubernetespythontypescript

Participation history

4 GSoC years

2026

9 projects

Official year page

OptimizationJob CRD for Hyperparameter Optimization in Kubeflow Katib

This project addresses limitations in Kubeflow Katib current Experiment CRD, which uses a generic and loosely typed interface for hyperparameter...

Project 5: Helm Charts for Kubeflow Pipelines and Katib — Danish Siddiqui

KFP and Katib users deploying via Helm currently have no upstream-supported path — charts exist but drift from Kustomize baselines silently, with no...

Project 6: MCP Server for Kubeflow SDK

The Kubeflow SDK gives AI practitioners a clean Python interface to submit, monitor, and manage distributed training jobs on Kubernetes via...

Agentic RAG on Kubeflow — Multi-Index Retrieval with Kagent & MCP

Kubeflow's documentation, GitHub issues, and platform code are spread across dozens of repositories with no unified search. This project evolves...

Kubeflow SDK/SparkClient - Batch Jobs, Observability & Production Readiness

The current Kubeflow SparkClient (KEP-107) provides a solid foundation for running interactive Spark workloads on Kubernetes, but it is still missing...

Platform Scalability and Security

Kubeflow's adoption at enterprise scale exposes critical bottlenecks in controller efficiency, security posture, and operational overhead. This...

End-to-End ARM64 Support & Validation on Kubeflow

ARM64 is becoming increasingly common with the rise of Apple Silicon and cloud instances like AWS Graviton, but Kubeflow still doesn’t run...

Project 11 - Composable Kale Notebooks with Visual Pipeline Editor

Kale compiles a single Jupyter notebook into a Kubeflow Pipeline by parsing cell tags, detecting data dependencies with PyFlakes, and generating KFP...

Dynamic LLM Trainer Framework for Kubeflow — TRL Backend with Pluggable Multi-Framework Support

Kubeflow Trainer V2 currently supports only TorchTune as its LLM fine-tuning backend. TorchTune stopped adding new features in July 2025, leaving...

2025

11 projects

Official year page

Project 7: GPU Testing for LLM Blueprints

This project aims to use self-hosted runners to run GPU-intensive tasks like LLM blueprint or (planned) AI Playground. The necessary infra is...

Project 5: JupyterLab Plugin for Kubeflow

This project addresses the issue of fragmented tools within the Kubeflow ecosystem, specifically Elyra, Kale, and Jupyter Scheduler, which...

Project3: Istio CNI and Ambient Mesh for Secure Kubeflow Deployments

This project aims to enhance Kubeflow's security posture by implementing Istio CNI by default and providing an optional Ambient Mesh configuration....

Batch Processing Gateway Integration

Jupyter Notebooks provide a powerful, interactive environment widely used by data scientists and ML engineers for exploratory analysis,...

Project 2: Kserve Models Web App

The KServe Models Web App is a crucial front-end interface for managing ML models deployed through KServe, a core component of Kubeflow. However, the...

Project 1: Kubeflow Platform Enhancements

This project aims to enhance Kubeflow’s infrastructure by productionizing the SeaweedFS proof of concept as a secure alternative to MinIO, improving...

Deploying Kubeflow with Helm

The proposed project aims to bridge a critical gap in the Kubeflow ecosystem by introducing robust Helm-based deployment support. While Kustomize has...

Project 8: Support JAX and TensorFlow Training Runtimes

This project implements key components of KEP-2170, introducing the Kubeflow Training V2 API. Specifically, it focuses on creating the...

Project 12: Empowering Kubeflow Documentation with LLMs

refer notion site for better view :...

Project 10: Support Volcano Scheduler in Kubeflow Trainer

This document outlines a proposal to support Volcano for gang-scheduling in Kubeflow Trainer, so as to provide users with more AI-specific scheduling...

Project 11: Support Postgres for Kubeflow Pipelines Backend

Kubeflow Pipelines (KFP) currently only supports MySQL/MariaDB databases, creating a significant limitation for many organizations that standardize...

2024

8 projects

Official year page

Push-based Metrics Collection for Katib (Project 6)

The project aims to provide a Python SDK API interface for users to push metrics to Katib DB directly. The current implementation of Metrics...

Automated Documentation Generation for Kubeflow Python SDKs Project Proposal

This proposal aims to automate the generation of comprehensive documentation for Kubeflow Python SDKs, specifically focusing on the Training Operator...

(Project Proposal) Project 8: Support various parameter distributions like log-uniform in Katib

The goal of this project is to enhance the Katib Experiment APIs to support various parameter distributions such as uniform, log-uniform, and...

Integrate JAX with Kubeflow Training Operator

The machine learning (ML) community is continually seeking ways to streamline the process of training and deploying models. JAX, with its...

Kubeflow Notebooks 2.0

This proposal aims to introduce a new Kubernetes controller for Kubeflow Notebooks. The implementation of this project will significantly enhance the...

LLM Hyperparameters Tuning API Proposal (Project 4)

This proposal aims to address the challenge of tuning hyperparameters for Large Language Models (LLMs) by developing a user-friendly high-level API....

Triage and Categorize Kubeflow GitHub Issues & PRs

The main goal of the project is to find a solution to categorize issues and PRs in the Kubeflow Github repository. There are a lot of issues and PRs...

Rootless Kubeflow Container Images(Istio Ambient Mesh)

My proposal aims to test and integrate the featues of Istio-CNI and Kubernetes' Pod Security Standards that are require to run Kubeflow rootless as...

2020

2 projects

Official year page

Build a UI to Deploy Managed Tensorboard Instances with Support for PVCs

Many Data Scientists have their data saved in a PersistentVolumeClaim accessed by their Notebooks. So it would be helpful and greatly improve...

Create a sample for notebook to Kubeflow deployment using TensorFlow 2.0 Keras

Kubernetes is already an industry-standard in managing cloud resources. Kubeflow is on its path to become an industry standard in managing machine...