Projects

### Overview: KB-NER is a knowledge-based system that achieved top results in the SemEval 2022 MultiCoNER shared task, ranking 10 out of 13 tracks. The system leverages a multilingual knowledge base built on Wikipedia to enhance the accuracy of complex named entity recognition. ### Key features: - **Knowledge Base Building**: KB-NER constructs a multilingual knowledge base using Wikipedia to provide contextual information for named entity recognition. - **Retrieval-based Data Augmentation**: The system uses retrieval-based data augmentation techniques to improve model performance. - **Context Processing**: KB-NER processes the context around named entities to enhance the accuracy of the recognition. - **Multi-stage Fine-tuning**: The model undergoes multiple stages of fine-tuning to improve its performance. - **Majority Voting Ensemble**: KB-NER employs a majority voting ensemble approach to improve accuracy. ### Tech stack: Python ### Why it matters: KB-NER's ability to leverage a multilingual knowledge base and advanced data augmentation techniques has significantly improved the accuracy of named entity recognition across various languages, making it a valuable tool for researchers and developers working on natural language processing tasks. ### Repository: Alibaba-NLP/KB-NER ### Primary language: Python ### GitHub stars: 185 ### GitHub description: Winner system (DAMO-NLP) of SemEval 2022 MultiCoNER shared task over 10 out of 13 tracks. [[Rankings]](https://multiconer.github.io

Overview: This Python wrapper provides access to Black Forest Labs' FLUX 3 Dev API, offering a faster, lower-cost variant of the FLUX 3 model. It includes Text-to-Image, Image-to-Image, Text-to-Video, and Image-to-Video functionalities, with the flagship Text-to-Image endpoint being live now. Key features: - Text-to-Image - Image-to-Image - Text-to-Video - Image-to-Video Tech stack: Python Why it matters: This SDK simplifies the integration of FLUX 3 Dev into projects, enabling rapid iteration and prototyping with reduced latency and cost compared to the full FLUX 3 model.

### Overview: MedXpertQA is a highly challenging and comprehensive benchmark designed to evaluate expert-level medical reasoning and understanding. It includes 4,460 questions spanning 17 specialties and 11 body systems, with two subsets: MedXpertQA Text for text medical evaluation and MedXpertQA MM for multimodal medical evaluation. This benchmark aims to provide a realistic and challenging environment for evaluating the performance of AI models in medical reasoning tasks. ### Key Features: - **Next-Generation Multimodal Medical Evaluation:** MedXpertQA MM introduces expert-level medical exam questions with diverse images and rich clinical information, including patient records and examination results, setting it apart from traditional medical multimodal benchmarks with simple QA pairs generated from image captions. - **Highly Challenging:** MedXpertQA introduces high-difficulty medical exam questions and applies rigorous filtering and augmentation, effectively addressing the insufficient difficulty of existing benchmarks like MedQA. The Text and MM subsets are currently the most challenging benchmarks in their respective fields. - **Clinical Relevance:** MedXpertQA incorporates specialty board questions to improve clinical relevance and comprehensiveness by collecting relevant medical data and integrating them into the evaluation process. ### Tech Stack: Python ### Why It Matters: MedXpertQA provides a realistic and challenging environment for evaluating AI models in medical reasoning tasks, making it an essential benchmark for researchers and developers in the field.

Overview: TriviumDB is a lightweight embedded database engine that seamlessly fuses vector search, graph, and document-oriented metadata within a single storage kernel. It is designed to be battle-tested in mission-critical, air-gapped environments, offering a highly efficient and secure solution for AI applications. Key features: - Vector search (Vector) - Graph (Graph) - Document-oriented metadata (Document) Tech stack: Rust Why it matters: TriviumDB simplifies the development of AI applications by providing a unified database solution that integrates vector search, graph processing, and document metadata. This eliminates the need for separate databases and simplifies the development process, making it an essential tool for AI developers looking to build robust and efficient applications.

Overview: This roadmap provides a beginner's path through the essential skills required for AI engineering in 2026, focusing on the shift from "train models" to "build reliable agent systems" and the three skills (RAG, Agents, and Evaluation) that appear in every job description. Key features: - Foundations - LLM Engineering - Agents - Evaluation - Production (LLMOps) Tech stack: None specified Why it matters: This roadmap is designed to help aspiring AI engineers navigate the rapidly evolving landscape of AI engineering in 2026, emphasizing the skills needed to build reliable agent systems and the importance of evaluation engineering. Repository: h9-tec/ai-roadmap-2026 GitHub stars: 178 GitHub description: AI Roadmap 2026: a beginner's path through Foundations, LLM Engineering, Agents, Evaluation, and Production, with verified free resources and a mini-project per layer.

Overview: This repository accompanies the ICML 2026 paper 'Meta Context Engineering via Agentic Skill Evolution'. It introduces MCE, a bi-level agentic framework that replaces rigid context engineering heuristics with learnable skills for optimal context optimization. Key features: - MCE formalizes context as a context function, combining static components and dynamic operators. - It includes a learnable skill for context optimization, improving performance across diverse domains. - MCE achieves significant efficiency gains over traditional methods, including 13.6x faster training and 4.8x fewer rollouts required. Tech stack: Python 3.11 Why it matters: MCE breaks free from the limitations of manually crafted context engineering methods, offering a more flexible and efficient approach to context optimization. Repository: metaevo-ai/meta-context-engineering GitHub stars: 183

### Overview: 给定 GPU 算力，一个模型能扛多少 QPS？这本手册详细解答了容量规划的三大核心问题，从单机三面墙到排队论 P95，从开环压测到 500 万 DAU 的采购决策，系统地拆解了容量规划的三层模型，提供完整代入算例和 SOP，帮助开发者解决实际落地中的各种问题。 ### Key features: - **物理层**: 理论上限多少？ - **系统层**: 单节点实际能服务多少？ - **排队层**: 真实流量下 P50/P95 多少？ ### Tech stack: 闭环压测、三面墙、排队论、开环压测 ### Why it matters: 系统性地解答了推理落地中绕不开的问题，帮助开发者准确规划推理能力，避免常见的错误方案，从而提升业务效率和成本效益。

### Overview Enthusiast is an open-source toolkit designed for building AI-powered agentic workflows. It includes tools like Retrieval-Augmented Generation (RAG), vector search, and workflow orchestrator, and is compatible with cloud-based models and self-hosted deployments. ### Key Features - **Retrieval-Augmented Generation (RAG)**: Enhances AI responses with relevant context from a database or knowledge base. - **Vector Search**: Quickly retrieves information from a large dataset. - **Workflow Orchestration**: Simplifies the creation and management of complex workflows. - **Customizable Agents**: Tailored to specific use cases, such as product discovery, user manual search, order intake, and catalog enrichment. ### Tech Stack Python/Django/PostgreSQL/React ### Why It Matters Enthusiast helps streamline AI development by reducing time, cost, and complexity. It provides validation and evaluation components to ensure responses are grounded in actual data, making it suitable for applications requiring high accuracy and reliability. ### Repository Repository: upsidelab/enthusiast GitHub Stars: 182 GitHub Description: Open source AI agent for e-commerce ### Primary Language Python ### Example Use Cases - **AI-Based Product Discovery**: Enables natural-language product search. - **User Manual Search**: Provides precise answers from technical documentation and manuals. - **Order Intake**: Converts unstructured inputs into e-commerce orders. - **Product Catalog Enrichment**: Extracts product descriptions, attributes, or translations from unstructured product sheets. ### Getting Started

### Overview: This repository is dedicated to the "Scaling Computer-Use Grounding via UI Decomposition and Synthesis" project, focusing on the development of a computer-use grounding technique using UI decomposition and synthesis. The project includes a benchmark dataset called OSWorld-G and a dataset pipeline called Jedi, which are essential for evaluating and improving the model's performance. ### Key Features: - **Benchmark Dataset (OSWorld-G)**: Contains original and refined instructions for computer-use grounding tasks, providing a comprehensive dataset for evaluation. - **Dataset Pipeline (Jedi)**: Includes the code and pipeline for collecting, processing, and refining data for computer-use grounding tasks. - **Model (Jedi-3B, Jedi-7B)**: Provides pre-trained models for computer-use grounding tasks, enhancing the model's performance with high-quality data. ### Tech Stack: TypeScript ### Why it Matters: This project offers valuable resources and benchmarks for researchers and developers interested in computer-use grounding techniques, providing a solid foundation for further advancements in this field.

### Overview: OSSSpeechKit is a Swift library designed for iOS developers to integrate speech synthesis and recognition functionalities using Apple's `AVFoundation` and `Speech` frameworks. ### Key Features: - **Speech Synthesis**: Supports multiple languages and provides a simple way to synthesize speech. - **Speech Recognition**: Enables real-time transcription of spoken text. - **Native Integration**: Built on top of Apple's native `AVFoundation` and `Speech` frameworks, ensuring seamless integration with iOS. ### Tech Stack: Swift 5 language mode or later, Xcode with Swift 6 package support. ### Why It Matters: - OSSSpeechKit simplifies the process of integrating speech capabilities into iOS applications, making it easier for developers to build voice-enabled features without the need for complex setup or third-party libraries. - It provides a robust solution for both speech synthesis and recognition, ensuring high-quality output and efficient performance. ### Repository: AppDevGuy/OSSSpeechKit - GitHub Stars: 184 - Primary Language: Swift - Installation: Swift Package Manager (only supported method) - Historical CocoaPods: 0.3.3 (no longer maintained) - CocoaPods: 1.0 and later (not published) - Usage: - **Speech Synthesis**: Supports multiple languages and provides a simple way to synthesize speech. - **Speech Recognition**: Enables real-time transcription of spoken text. - **Languages**: Supports over 100 languages. - **Voices**: Offers a variety of voices for different accents and dialects

### Overview: UHGEval is a comprehensive benchmarking framework designed to evaluate the hallucination capabilities of Chinese large language models. It includes an unconstrained hallucination evaluation benchmark, Eval Suite, and supports other benchmarks like HalluQA and HaluEval. ### Key features: - **UHGEval**: An unconstrained hallucination evaluation benchmark. - **Eval Suite**: A user-friendly evaluation framework for hallucination tasks. - **Supports**: Other benchmarks like HalluQA and HaluEval. ### Tech stack: Python ### Why it matters: UHGEval provides a robust and user-friendly platform for evaluating the hallucination capabilities of Chinese large language models, making it essential for researchers and developers in the field of natural language processing.

### Overview: ChaKt is a multiplatform chat-prompt based app built using Kotlin and Compose Multiplatform. It leverages Google's Generative AI Multiplatform SDK to generate content, showcasing the use of [Generative AI SDK for Kotlin Multiplatform](https://github.com/PatilShreyas/generative-ai-kmp). ### Key Features: - 📱 Mobile: Available for Android and iOS. - 🖥️ Desktop: Compatible with Windows, macOS, and Linux. - 🌎 Web: Accessible via a web interface. - 📂 Generative AI: Powered by Google's Generative AI models. ### Tech Stack: Kotlin and Compose Multiplatform ### Why it Matters: ChaKt demonstrates the integration of Generative AI with Kotlin Multiplatform, highlighting the potential for creating interactive and intelligent chat experiences across multiple platforms. ### GitHub Stars: 184 ### Repository: PatilShreyas/ChaKt-KMP

### Overview: This repository contains a collection of the best ML and AI news every week, covering research, news, resources, perspectives, and memes. It serves as a valuable resource for staying updated on the latest developments in artificial intelligence and machine learning. ### Key Features: - **Research**: The most important published research of the week. - **News**: The most important news related to companies, institutions, and more. - **Resources**: Released resources for artificial intelligence and machine learning. - **Perspectives**: A collection of deep and informative articles about open questions in artificial intelligence. - **Meme**: A meme to start the week well. ### Tech Stack: Unknown ### Why It Matters: This repository is a curated collection of the best news and resources in the field of artificial intelligence, making it a valuable tool for staying informed and up-to-date with the latest developments in the field.

Overview: This project is a UIButton subclass that includes features for recording audio, speech recognition, and displaying a waveform view similar to Siri. It utilizes Swift and is compatible with iOS. Key features: - Recording audio with push-to-talk functionality - Speech recognition using Siri-style waveform view - Customizable configuration for UIButton and SFWaveformView Tech stack: Swift Why it matters: This project is useful for developers looking to integrate advanced audio and speech recognition features into their iOS apps, providing a user-friendly interface for voice commands and recordings. Repository: alexruperez/SpeechRecognizerButton Primary language: Swift GitHub stars: 184 GitHub description: UIButton subclass with push to talk recording, speech recognition and Siri-style waveform view.

### Overview: A GStreamer plugin that wraps Kaldi's SingleUtteranceNnet2Decoder, enabling online processing of iVector-adapted DNN acoustic models. This plugin requires iVector adaptation and supports various features like phone and word alignments, dropout, and batch normalization. ### Key Features: - **Online Processing**: Automatically adapts iVectors to the current audio stream. - **Phone Confidence Scores**: Inclusion with phone alignments. - **Word Confidences**: Confidence scores for word alignments. - **Structured Recognition Results**: Outputs structured recognition results including n-best results, phone alignments, and possibly other metrics. - **Model Compatibility**: Supports FST, acoustic models, and big language models. - **Threaded Decoder**: Option to use a threaded decoder for improved performance. ### Tech Stack: C++ ### Why it Matters: This plugin simplifies the process of online speech recognition by providing a straightforward interface to Kaldi's advanced neural network decoders, making it easier to integrate into existing GStreamer pipelines.

**Overview:** Emote Portrait Alive aims to reverse engineer code from a white paper using AI techniques. The project is currently abandoned, focusing on the development of a model for emulating code-based emotes. **Key Features:** - Emote Portrait Alive: A project to reverse engineer code from a white paper using AI. - AniPortrait: Achieves near EMO-like results using the Vid2vid method. - Audio Attention: Exploring the use of audio attention for facial expressions. - Face Locator: A module that locates the face region in an image. - Speed Encoder: Extracts speed embeddings from audio waveforms. **Tech Stack:** Python **Why It Matters:** The project has the potential to revolutionize the way emotes are generated, offering a more realistic and dynamic representation of code-based expressions. **Repository:** johndpope/Emote-hack **GitHub Stars:** 184 **GitHub Description:** Emote Portrait Alive - using AI to reverse engineer code from white paper. (abandoned) **Repository:** johndpope/Emote-hack **Primary Language:** Python **GitHub Description:** Emote Portrait Alive - using AI to reverse engineer code from white paper. (abandoned) **Repository:** johndpope/Emote-hack **Primary Language:** Python **GitHub Description:** Emote Portrait Alive - using AI to reverse engineer code from white paper. (abandoned) **Repository:** johndpope/Emote-hack **Primary Language:** Python **

Overview: Speechly is a comprehensive platform offering Automatic Speech Recognition (ASR) and Natural Language Understanding (NLU) tools, SDKs, and APIs. It provides easy integration for web applications, supporting on-device, on-premise, and cloud deployment options. Key features: - Speechly Client Libraries: JavaScript and React APIs for natural spoken language understanding. - Demos: Source code for web-based demos like moderation and transcription. - Examples: Practical projects demonstrating Speechly’s capabilities. Tech stack: TypeScript Why it matters: Speechly simplifies the integration of speech recognition and language understanding into web applications, making it accessible for developers without requiring deep expertise in AI or backend development. Repository: speechly/speechly GitHub stars: 184 GitHub description: Client libraries, examples, and demos of Speechly API for the Web.

Overview: This project provides a versatile server that can interface with various OpenAI API clients, including Ollama, vLLM, and even OpenAI itself, all through the Responses API. It supports stateful chat, tool calls, and future features like file search and code interpretation, making it suitable for developers looking to experiment with self-hosted LLMs while maintaining compatibility with OpenAI's API. Key features: - MCP (Model-Client-Protocol) support - Stateful chat handling - Tool call support - Automatic and manual tests - Docker image build - PyPI release - Persistent state storage - CLI validation - Hosted tools: MCPs support, web search, file upload & search, code interpreter, computer use APIs Tech stack: Python Why it matters: This project simplifies the integration of OpenAI's Responses API with various backend AI models, making it easier for developers to experiment with self-hosted LLMs while maintaining compatibility with OpenAI's API. It is designed for innovators, researchers, and OSS enthusiasts, offering enterprise-ready scalability, reliability, and security for production workloads. Repository: teabranch/open-responses-server

Overview: This project is a simple API server for the LLaVA model implementation, LLaVA server (llama.cpp), developed by Bart Trzynadlowski in 2023. Key features: - Starts a server on `localhost:8080` for interaction with the LLaVA model. - Accepts user prompts, image data, and system prompts via the `/llava` endpoint. - Supports HTTP logging for better debugging. Tech stack: C++ Why it matters: This server provides a convenient way to interact with the LLaVA model, making it easier to integrate the model into various applications or environments.

### Overview: EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling ### Key features: - Promotes equivariance in the latent space of pretrained autoencoders. - Accelerates generative model training. - Improves performance by structuring the latent distribution. ### Tech stack: Python ### Why it matters: EQ-VAE introduces a regularization objective that encourages equivariance in the latent space of pretrained autoencoders, leading to a more structured latent distribution that accelerates generative model training and improves performance.