Directory
Explore AI Infrastructure & MLOps resources in AI & Machine Learning.
Hopsworks is a platform for building, deploying, and scaling production machine learning systems. It combines a feature store with MLOps tooling, using dual online and offline storage to manage, reuse, and serve features. The platform covers the ML lifecycle: train, deploy, and monitor. An O'Reilly book covers building batch, real-time, and LLM systems with feature stores.
A 2003 book by David J.C. MacKay published by Cambridge University Press, covering information theory, inference, and learning algorithms. The site offers corrections, software, and downloadable figures. It lists a price of £35.00 or $60.00.
An archived educational course repository for learning about LLMs, LLMOps and vector databases by designing, training and deploying a real-time financial advisor LLM system. It includes source code, video and reading materials, and covers a training pipeline using QLoRA and a streaming pipeline ingesting financial news. The repository was archived by its owner and is read-only.
Build multimodal AI services via cloud native technologies · Model Serving · Generative AI · Neural Search · Cloud Native.
Kaito is the Kubernetes AI Toolchain Operator, an operator suite that automates LLM inference, fine-tuning and RAG engine deployment in a Kubernetes cluster. It simplifies the CRD API and provides optimized preset configurations for inference engine scheduling parameters. Workspaces are the basic CRD for managing inference and tuning workloads, and all vLLM-supported models can be run.
Kubeflow Katib is a Kubernetes-native project for automated machine learning (AutoML). It supports hyperparameter tuning, early stopping and neural architecture search, and is agnostic to machine learning frameworks, natively supporting frameworks such as TensorFlow, PyTorch and XGBoost. Training jobs can use any Kubernetes Custom Resources, with out-of-box support for Kubeflow Training Operator, Argo Workflows and Tekton Pipelines. A Python SDK simplifies creating tuning jobs.
Respan is an AI gateway that routes requests to over 1,000 models through one API endpoint, with automatic fallbacks, response caching, and spending limits. It also provides built-in observability for tracing LLM calls, tool calls, and agent steps, plus automated evaluations and security testing for masking PII and red-teaming agents before launch.
Kubernetes custom resource definition for serving ML models on arbitrary frameworks.
Kitaru is an open-source, self-hosted tool providing replay-based evaluations for AI agents. It records or imports production runs as sessions and replays them against a changed model, prompt, or code, with tool calls answered from the recording. It is available for Python and TypeScript and comes from the team behind ZenML.
Open source MLOps project that eases model handoffs between data scientist and DevOps.
Knowledge3D (K3D) is a sovereign GPU-native reference implementation for the W3C PM-KR specification. It aims to unify web knowledge representation by allowing humans and AI to consume the same procedural knowledge from a single source, addressing fragmentation across fonts, embeddings, metadata, and renderings.
KServe is a standardized distributed generative and predictive AI inference platform for scalable, multi-framework deployment on Kubernetes. It is a Cloud Native Computing Foundation incubating project. The platform supports backends such as vLLM and llm-d for serving LLMs, and multi-framework predictive serving. Standalone installation on standard Kubernetes is described as more lightweight but without canary deployment or request-based autoscaling with scale-to-zero.
Kubeflow is described as the cloud native AI platform, composed of modular open source projects forming a Kubernetes-native stack for data and AI workloads. Its subprojects include Kubeflow Pipelines for building and deploying portable machine learning workflows, Kubeflow Trainer for distributed AI training and LLM fine-tuning, Kubeflow Katib for automated machine learning, Kubeflow Notebooks for interactive development environments, and Kubeflow Hub for indexing models and metadata. It is a Cloud Native Computing Foundation project.
KubeStellar Console is an open-source multi-cluster Kubernetes management dashboard. It allows users to monitor clusters, deploy workloads, manage GPU resources, and troubleshoot with AI-powered missions across all your Kubernetes environments.
Laminar is an open-source observability platform purpose-built for AI agents. It provides OpenTelemetry-native tracing with a one-line SDK setup, plain-English signals that alert in Slack, and an SDK and CLI for running evals. It is written in Rust with 20x trace compression, and can be used via a managed platform or self-hosted with Docker Compose.
LanceDB is a developer-friendly open-source embedded retrieval library for multimodal AI. It emphasizes searching more and managing less. The project tagline is 'Search More; Manage Less.' It is available on GitHub.
LangWatch is a platform for evaluating large language models and testing AI agents. The open-source project is hosted on the GitHub platform under the langwatch organization as the langwatch/langwatch repository.
Latitude is an open-source observability tool for AI agents. It helps identify failures, dispatch coding agents to fix issues, and verify fixes using real traces. The platform is available on GitHub.
layer is a command-line tool for neural network inference, following the Unix philosophy. It enables users to perform inference from the command line. The project is open-source and hosted on GitHub.
libonnx is a lightweight, portable pure C99 ONNX inference engine for embedded devices with hardware acceleration support. Its .c and .h files can be dropped into a project and compiled with it. Models are loaded from a file into a context, and inference runs on input tensors via onnx_run. It is based on ONNX version v1.17.0 with opset 24 support, and some operators are not implemented.
The Libra Toolkit is a collection of algorithms for learning and inference with discrete probabilistic models, including Bayesian networks, Markov networks, dependency networks, sum-product networks and arithmetic circuits. Each algorithm is implemented as a command-line program with consistent options and file formats, and all methods are written in OCaml.
LiteLLM is an open-source library that provides a single, unified interface to call over 100 LLMs using the OpenAI format. It includes a Python SDK whose completion() function uses the same arguments across providers and returns responses in the OpenAI Chat Completions format, plus a self-hosted gateway with virtual keys, cost tracking and an admin UI.
Lunary is an observability and prompt management platform for LLM-based applications. It provides chatbot analytics, agent tracing, prompt templates, topic classification, PII masking and feedback tracking. A self-hostable option is offered. The Team cloud plan costs $20 per user per month, billed monthly, and payment is required to activate a workspace.
Mastra is a TypeScript AI framework for building agents and AI applications. It includes built-in memory, tool support, MCP compatibility, and observability features to support development from prototype to production. The framework also offers workflow tools, agent factory functionality, cloud deployment servers, and an agent builder for evaluation and iteration.
