Projects

Overview: The MultiMind SDK is a comprehensive AI framework that simplifies the integration of various models and provides built-in compliance features, making it easy to deploy AI solutions without worrying about compliance. Key features: - A single API for accessing multiple models and local models - Built-in GDPR and HIPAA support - Compliance dashboards and policies as a first-class module - Governance of existing models and frameworks - Adaptive routing for model switching - RAG (Relevance-Aware Generation) that works with FAISS and Chroma - Runs anywhere, including cloud, on-prem, and air-gapped setups Tech stack: Python Why it matters: The MultiMind SDK simplifies the integration of various AI models, providing a unified interface and built-in compliance features, making it easier to deploy AI solutions without the hassle of compliance management.

Overview: Branches is an AI tool designed for researchers and developers to prototype advanced algorithms for LLM reasoning and planning, particularly focusing on tree-based reasoning and code generation. Key features: - Automated Tree Expansion - Pre-loaded Prompts - Code Interpretation - Scoring Mechanism - Interactive Visualization Tech stack: TypeScript Why it matters: Branches provides a user-friendly interface for exploring complex coding challenges and strategic problem-solving through advanced AI reasoning processes. Repository: normal-computing/branches Primary language: TypeScript GitHub stars: 97

Overview: Image Context Runtime for Codex is an experimental open-source plugin and local runtime that keeps Codex workflows responsive and context-bounded by generating and inspecting image jobs as persisted artifacts. Key features: - Image generation jobs - Image inspection jobs - Durable status and text handoffs - Idempotent submission - Restart reconciliation and explicit recovery - Offline deterministic mock provider - Optional OpenAI Image API and Responses API provider - Strict public-result budgets with no MCP image, audio, or resource blocks Tech stack: JavaScript Why it matters: This project reduces one source of context pressure in Codex workflows, ensuring that image-heavy tasks remain efficient and manageable.

Overview: This repository contains practical examples, experiments, and implementations for the "Applied AI" book by Sinan Ozdemir. It covers foundational concepts in LLMs, embeddings, agents, and more, with a focus on building reliable, scalable, and efficient AI systems. Key features: - Code examples - Case studies on various AI topics - Prerequisites for Python 3.8 or higher and common dependencies - Environment variables for handling sensitive information Tech stack: Jupyter Notebook Why it matters: This repository serves as a practical resource for developers, data scientists, and founders looking to understand and implement modern AI systems, providing a solid foundation and real-world applications. Repository: sinanuozdemir/building-agentic-ai GitHub stars: 97 GitHub description: Code repository for "Building Agentic AI" by Sinan Ozdemir. Practical examples, experiments, and implementations covering LLMs, embeddings, agents, RAG, fine-tuning, and optimization. Primary language: Jupyter Notebook GitHub description: Code repository for "Building Agentic AI" by Sinan Ozdemir. Practical examples, experiments, and implementations covering LLMs, embeddings, agents, RAG, fine-tuning, and optimization.

Overview: This project implements a PyTorch-based solution for Automatic Pronunciation Error Detection (APED) using Noisy Student Training (NST) for Automatic Speech Recognition (ASR). The system uses pre-trained Wav2Vec 2.0 and self-training techniques to improve phoneme recognition accuracy. Key features: - Implementing Noisy Student Training for ASR and APED - Utilizing PyTorch for the project - Training on LibriSpeech and Libri-Light datasets - Detecting phoneme sequence errors through Longest Common Subsequence (LCS) Tech stack: PyTorch Why it matters: This project offers a practical approach to improving pronunciation accuracy in speech recognition tasks, which is crucial for language learning and education.

Overview: CommonScenes is an innovative method for generating 3D indoor scenes using scene graphs and diffusion models, showcasing its ability to create entire scenes from scene graphs, including layout and 3D geometries. Key features: - Utilizes scene graphs to generate 3D scenes - Employs diffusion models for scene generation - Supports layout and 3D geometry in generated scenes Tech stack: Python Why it matters: CommonScenes provides a novel approach to generating realistic 3D indoor scenes, offering a holistic solution that integrates scene layout and 3D geometry, making it valuable for applications requiring detailed and contextually relevant 3D environments.

Overview: SciLink is an AI-powered platform designed to automate scientific research workflows, leveraging large language models and domain-specific tools to streamline experimental design, data analysis, and optimization. Key features: - Planning Agents: Generate hypotheses, optimize experiments, and plan experiments. - Analysis Agents: Perform multi-modal data analysis across various modalities. - Simulation Agents: Conduct computational modeling for DFT calculations and classical MD simulations. Tech stack: Python Why it matters: SciLink automates the entire scientific research cycle, from experimental design to data analysis, making researchers more efficient and reducing the time spent on manual tasks. Repository: ziatdinovmax/SciLink Primary language: Python GitHub stars: 97 GitHub description: LLM-powered agents for scientific research automation

### Overview: ESEILANE is a high-performance Knowledge Graph engine designed specifically for Large Language Models (LLMs), GraphRAG, and AI-native applications. Leveraging sparse matrix algebra and linear algebra-based query execution, ESEILANE delivers sub-millisecond graph traversals at scale, making it ideal for applications requiring rapid and accurate knowledge retrieval. ### Key Features: - **Ultra-Low Latency**: Sub-millisecond response times powered by GraphBLAS sparse matrix algebra. - **GraphRAG Native**: First-class integration with LLMs, reducing hallucinations and improving AI accuracy. - **Property Graph Model**: Nodes and relationships with rich attributes, fully supported by OpenCypher. - **Horizontal Scale**: Pay-as-you-grow distributed architecture with zero-overhead multi-tenancy. - **Enterprise Ready**: Role-based access control, audit logging, encryption at rest and in transit. - **Rust Core**: Next-gen engine rewritten in Rust for maximum performance and memory safety. ### Why It Matters: - **Ultra-Low Latency**: Delivers sub-millisecond graph traversals, essential for applications requiring real-time responses. - **GraphRAG Native**: Seamless integration with LLMs, reducing hallucinations and improving AI accuracy. - **Property Graph Model**: Nodes and relationships with rich attributes, enabling advanced querying capabilities. - **Horizontal Scale**: Pay-as-you-grow distributed architecture, allowing for zero-overhead multi-tenancy. - **Enterprise Ready**: Robust security features, including role-based access control, audit logging, and encryption. - **Rust Core**: High

Overview: This project is a comprehensive guide for AI development in 2025, focusing on Spring Boot 3, Java 21, and Spring AI. It includes AI model integration, core features of Spring AI, Prompt engineering, RAG techniques, vector databases, tool calling, MCP model context protocol, AI agent development, and Cursor AI tools. The goal is to equip programmers with essential AI skills and enhance their job prospects in the AI-driven job market. Key features: - AI model integration - Spring AI core features - Prompt engineering - RAG (Retrieval-Augmented Generation) - Vector databases - Tool calling - MCP (Model Context Protocol) - AI agent development - Cursor AI tools Tech stack: Java 21, Spring Boot 3, Spring AI Why it matters: This project provides a structured tutorial for programmers to learn essential AI skills, making them competitive in the AI-driven job market and enhancing their resume and job prospects.

Overview: PROMETHEVS is a powerful, open-source engine that runs the full DeepSeek-V4.1-Flash-1017 model on a gaming PC with 64 GB RAM and a 12 GB GPU. It compresses and corrects the model, achieving high accuracy and speed. Key features: - Runs the full DeepSeek-V4.1-Flash-1017 model - Compress-then-correct method for high accuracy - 98.3% accuracy for 960 audited words - Up to 7.4 tokens/s on code prompts - Fast follow-up responses - Compatible with OpenAI-compatible endpoints - Runs on your PC, private and free Tech stack: Python, JavaScript, PowerShell, and various libraries for compress-then-correct method Why it matters: PROMETHEVS offers a way to run top-tier AI models on less powerful hardware, making it accessible to a wider audience. Its methods ensure high accuracy and fast performance, making it valuable for applications requiring high precision and speed. Repository: Magnvs7/Promethevs GitHub stars: 69 GitHub description: PROMETHEVS: run the full DeepSeek-V4.1-Flash-1017 on a gaming PC with 64 GB RAM and a 12 GB GPU. PROMETHEVS engine (compress-then-correct), OpenAI-compatible local server, report and logs.

### Overview: This repository is a curated list of self-hosted open-source AI projects, including local LLMs, chat interfaces, agents, MCP servers, RAG, speech and image generation. The projects are health-checked weekly, and anything that is no longer maintained is archived in the [UNMAINTAINED.md](UNMAINTAINED.md) file. ### Key features: - **Local LLMs**: Projects that allow running large language models locally. - **Chat Interfaces**: Tools for building conversational AI chatbots and agents. - **Agents and Automation**: Frameworks and libraries for building multi-agent systems and orchestrating AI teams. - **Coding Assistants**: Platforms for building and self-hosting conversational AI chatbots and agents. - **Document AI and OCR**: Tools for document analysis and optical character recognition. - **Fine-Tuning and Training**: Resources for fine-tuning and training AI models. - **Home and Personal AI**: Projects for creating AI assistants for personal use. - **Image and Video Generation**: Tools for generating images and videos using AI. - **LLM Runtimes and Inference**: Platforms for running large language models. - **MCP Servers and Gateways**: Servers and gateways for managing and orchestrating AI agents. - **Model Gateways and Routers**: Tools for routing and directing AI traffic. - **Observability and Evals**: Tools for monitoring and evaluating AI models. - **RAG and Vector Databases**: Resources for using retrieval augmented generation (RAG) and vector databases. - **Search and Knowledge**: Tools

### Overview: This project serves as a comprehensive guide for AI development in 2025, focusing on creating the AI Love Master application and the ReAct mode autonomous planning agent, YuManus. It covers core knowledge on AI model integration, Spring AI features, Prompt engineering, RAG techniques, vector databases, tool calling, MCP protocol, AI agent development, and Cursor AI tools. ### Key Features: - AI model integration - Spring AI core features - Prompt engineering - RAG techniques - Vector databases - Tool calling - MCP protocol - AI agent development - Cursor AI tools ### Tech Stack: Java 21, Spring Boot 3, Spring AI ### Why it Matters: This project aims to equip programmers with essential AI skills, making them competitive in the AI-driven job market. By learning about AI model integration, Spring AI features, Prompt engineering, RAG techniques, vector databases, tool calling, MCP protocol, AI agent development, and Cursor AI tools, programmers can enhance their resume and job prospects in the AI-driven job market. ### GitHub Stars: 96 ### Repository: hfgwygey/yu-ai-agent

### Overview: Stable Diffusion Photopea is a plugin for PhotoPea that leverages the A1111 API to integrate Stable Diffusion, allowing users to create real-time and interactive art directly within the photo editor. ### Key Features: - **Real-time Rendering**: Utilizes the Latent Consistency Model (LCM) for faster inference, enabling immediate rendering of the canvas. - **Configurable LCM**: Supports multiple LCM configurations for enhanced control and performance. - **Real-time Selection**: Renders only the selected area, enhancing efficiency. - **ControlNet Scribble**: Enables the use of ControlNet's scribble model, allowing for more detailed and precise painting. - **Reference Range Selection**: Automatically references a limited range of surroundings during image processing, addressing pain points in large image inpaints. ### Tech Stack: Vue ### Why It Matters: Stable Diffusion Photopea simplifies the process of creating art within PhotoPea, providing a seamless and efficient workflow. The integration of LCM and ControlNet enhances the capabilities of the plugin, making it a valuable tool for both beginners and experienced users.

### Overview: Black Mirror is a voice-controlled smart mirror powered by Raspberry Pi 3 and Android Things, designed to display and hide widgets such as weather, news, time, and calendar. ### Key Features: - **Voice Control**: Uses a Raspberry Pi 3 to interpret voice commands. - **Widget Control**: Can show or hide widgets like weather, news, time, and calendar. - **Language Support**: Currently supports Polish commands. ### Tech Stack: Java ### Why it Matters: Black Mirror is a practical and user-friendly solution for those looking to add a smart mirror feature to their Android devices. It's designed to be simple and intuitive, making it accessible to a wide range of users.

### Overview: Face2Diffusion is a cutting-edge research project that introduces a novel method for fast and editable face personalization. This project was presented at CVPR 2024 and offers a comprehensive solution for transforming faces in real-time applications. ### Key Features: - **Face2Diffusion**: A PyTorch implementation for the CVPR 2024 paper. - **Fast Personalization**: The system allows for quick and efficient face personalization without significant computational overhead. - **Editable Faces**: Users can easily modify and edit their faces in real-time using the provided tools. ### Tech Stack: - **Language**: Jupyter Notebook - **GitHub Stars**: 96 ### Why It Matters: Face2Diffusion provides a practical and user-friendly solution for face personalization, making it accessible for a wide range of applications, including social media, virtual assistants, and more. ### Repository: - **Repository**: mapooon/Face2Diffusion - **Primary Language**: Jupyter Notebook - **GitHub Stars**: 96 - **GitHub Description**: [CVPR 2024] Face2Diffusion for Fast and Editable Face Personalization ### Changelog: - **2024/03/28**: Released a demo code on Google Colab. - **2024/03/15**: Released the full inference pipeline. - **2024/03/11**: Released this repository and demo code. ### Demo on Google Colab: [![**face2

### Overview: Orka is an AI orchestration system designed to transform Large Language Models (LLMs) into composable agents capable of reasoning, fact-checking, and constructing answers with transparent traceability. It is a modular toolkit that allows developers to build complex AI workflows without the need for deep technical expertise. ### Key Features: - **Orchestrated Workflow Management:** Orka enables developers to create and manage complex AI workflows by orchestrating multiple agents. - **Transparent Traceability:** Each step in the workflow is logged, ensuring that the reasoning process is traceable and understandable. - **Fact-Checking and Reasoning:** Agents can fact-check information and reason through complex problems, providing accurate and contextually relevant answers. - **Scalability and Flexibility:** Orka supports various programming languages and can be easily integrated into existing projects. ### Tech Stack: - Python ### Why It Matters: Orka is crucial for developers looking to build AI-driven applications that can handle complex reasoning tasks. It provides a framework that simplifies the process of integrating AI into existing systems, making it easier for developers to implement AI solutions without extensive technical knowledge.

### Overview: This ROS 2 project provides a set of packages to integrate Whisper with SileroVAD, enabling speech-to-text functionality using the GGML Whisper model. ### Key Features: - Whisper integration for speech-to-text - Voice Activity Detection (VAD) using SileroVAD - ROS 2 compatible packages ### Tech Stack: C++ ### Why It Matters: This project simplifies the integration of Whisper with ROS 2, making it easier for developers to implement speech recognition in their projects.

Overview: Image Studio 是一个集图解、绘画、编辑和视频生成于一体的AI视觉创作套件，支持Vercel一键部署。 Key features: - 图解百科 (Illustrated Wiki) - 连环画本 (Comic Strip) - 无限画布 (Infinite Canvas) - 以文生图 (Text-to-Image) - 以图生图 (Image-to-Image) - 图生视频 (Image-to-Video) - 历史记录 (History) - 一键部署 (Deploy) Tech stack: TypeScript Why it matters: Image Studio 是一个功能丰富、流程顺畅的视觉内容创作平台，集成了Google最新的人工智能模型，旨在帮助用户轻松创作出高质量的视觉作品。

### Overview: Milvus insight is an intuitive GUI tool designed for managing and interacting with Milvus databases. ### Key features: - Manage collections/partitions - Manage indexes - Basic statistics overview - Load/release collections for search - Insert entities - Vector search with advanced filter - Milvus system view (TBD) - Data view (TBD) - View milvus node configuration (TBD) - Vector Visualization (TBD) - More... ### Tech stack: TypeScript ### Why it matters: Milvus insight simplifies the management and interaction with Milvus databases, making it easier for developers to work with vector databases.

### Overview: Simplifine is a powerful, easy-to-use open-source tool for fine-tuning Large Language Models (LLMs) on any dataset or model. It simplifies the process by handling all infrastructure, job management, cloud storage, and inference, making it perfect for developers and researchers. ### Key features: - **Easy Cloud-Based LLM Finetuning:** Fine-tune any LLM with just one command. - **Seamless Cloud Integration:** Automatically manage the downloading, storing, and running of models directly from the cloud. - **Built-in AI Assistance:** Get help with hyperparameter selection, synthetic dataset generation, and data quality checks. - **On-Device to Cloud Switching:** Add a simple decorator to transition from local to cloud-based training. - **Auto-Optimization:** Automatically optimizes model and data parallelization through Deepspeed and FDSP. - **Custom Evaluation Support:** Use the built-in LLM for evaluations functions or import your own custom evaluation metrics. - **Community Support:** Asking any support questions on the Simplifine Community Discord. - **Trusted by Leading Institutions:** Research labs at the University of Oxford rely on Simplifine for their LLM finetuning needs. ### Tech stack: Python ### Why it matters: Simplifine streamlines the process of fine-tuning LLMs, making it accessible and efficient for developers and researchers. Its seamless integration with popular optimization frameworks and its built-in AI assistance make it a valuable tool in the field of natural language processing and language modeling.