Projects

Overview: Prosodic is a Python library and web app designed for metrical-phonological analysis of poetry, parsing text into a linguistic hierarchy and identifying stress patterns, foot/syllable schemes, and rhyme schemes. It supports English and Finnish, with flexible language support. Key features: - Parses text into a linguistic hierarchy - Runs a constraint-satisfaction metrical parser - Identifies stress patterns (iambic, trochaic, anapestic, dactylic) - Identifies foot/syllable schemes - Identifies named rhyme schemes (sonnet variants, couplet, ballad, etc.) Tech stack: Prosodic is written in Python and uses espeak for phonemizing words not in the CMU dictionary. Why it matters: Prosodic is useful for analyzing the metrical and phonological structures of poetry, providing insights into the rhythm and rhyme patterns of texts. It can be used to classify poem forms and identify common rhyme schemes, making it valuable for both academic and creative writing projects.

Overview: This repository is a curated list of open-source AI resources, focusing on inference engines, language models, embedding models, image models, audio models, and tools. It aims to provide resources that meet the Open Source Initiative (OSI) definition of open source. Key features: - Koboldcpp: A swiss army knife of LLM inference engines that supports multiple API endpoints. - Llama.cpp: A barebones LLM inference engine with additional options for controlling LLM output. - Ollama: An easy-to-use LLM inference engine offering an OpenAI-compatible API server. - OpenedAI Speech: A great OpenAI-compatible API TTS server compatible with Piper TTS, optimized for CPU performance. - DeepSeek-R1: A language model suited for reasoning and logic-related tasks. - Granite 3.1: A language model suitable for RAG and deployment on low-end consumer hardware. - InternVL 2.5: A language model for vision-related tasks, scoring better on benchmarks compared to Qwen2-VL. - OLMo 2: A completely open language model with all parts of the model training process open-source, scoring well on benchmarks. - Phi-4: A great language model that is SOTA in various benchmarks for a medium-sized language model, suitable for instruction-following and logical tasks. - Qwen 2.5: Best in-class language models that excel in benchmarks for math, etc., but may have censorship on certain topics. - Qwen 2.5 Coder: Best in-class language models that excel

### Overview: This is a comprehensive open-source repository for teaching Data Science and Artificial Intelligence, providing course notes, slides, and code in LaTeX. The repository includes full courses, workshops, seminars, and runnable projects, covering a wide range of topics from machine learning to natural language processing and graph neural networks. ### Key Features: - **Courses**: Full curriculum (Machine Learning, Maths for ML, Python, Deep Learning, Generative AI) with 1-2 weeks of content. - **Workshops**: Focus on one topic, typically 1-2 days. - **Seminars**: Focused sessions, typically 1-2 hours. - **Code Projects**: Runnable Python/notebook projects for various AI and ML tasks. ### Tech Stack: - **LaTeX**: All content is prepared and compiled using LaTeX. ### Why It Matters: The repository is designed to democratize access to high-quality, open-source educational materials, making it easier for anyone to learn and teach Data Science and AI. The open-source nature ensures that the content can be freely used, modified, and shared by the community. ### Repository: - **Repository**: `yogeshhk/TeachingDataScience` - **GitHub Stars**: 305 - **GitHub Description**: Open-sourced course notes for Artificial Intelligence and Data Science related topics, prepared in LaTeX. ### Getting Started: To get started, clone the repository and navigate to the `LaTeX` directory. Compile the presentation slides using `texify -cp Main_Seminar_ML_Intro_Presentation.tex`. This

### Overview: Flappy is a production-ready Language Language Model (LLM) Application/Agent SDK designed to simplify AI integration in your projects. It is an easy-to-use, universally compatible, and production-ready solution that brings the power of AI to developers regardless of their preferred programming language. ### Key Features: - **Ease of Use**: Flappy is designed to be as user-friendly as CRUD application development, minimizing the learning curve for developers new to AI. - **Production-Ready**: Beyond research, Flappy is a robust SDK that balances cost-efficiency and sandbox security to provide a stable platform for commercial environments. - **Language-Agnostic**: Flappy integrates seamlessly with any programming language, eliminating the need for Python unless explicitly required by your application. ### Tech Stack: Rust ### Why It Matters: Flappy simplifies the integration of AI into projects, making it accessible to developers of all backgrounds. It is designed to be production-ready, ensuring stability and security in commercial environments.

### Overview: Rasa NLU GQ is a powerful tool that converts natural language into structured data, making it easier to extract and understand information from text. It supports Chinese and allows for the creation of various models tailored to specific scenarios and tasks. ### Key features: - **Entity Recognition**: Enhanced model for identifying entities such as names, locations, and institutions. - **Jieba Tokenizer**: Improved word segmentation using Jieba, a popular Chinese tokenizer. - **CRF Entity Extractor**: Enhanced entity extraction for better identification of entities. - **BERT Model Featurizer**: Extracts word vectors using BERT, enhancing the model's ability to understand context. - **Intent Classifier with TensorFlow**: Uses TensorFlow for advanced classification, making it more efficient and flexible. ### Tech stack: Python ### Why it matters: Rasa NLU GQ simplifies the process of converting natural language into structured data, making it easier to extract and understand information from text. It supports Chinese and allows for the creation of various models tailored to specific scenarios and tasks, making it a valuable tool for developers working on natural language processing projects.

Overview: The Naomi Project is an open-source platform designed for developing always-on, voice-controlled applications. Key features: - Control Anything - Always Listening - 100% Open Source - Cross Platform (Raspberry Pi and Linux) - Thorough Documentation Tech stack: Python Why it matters: The Naomi Project provides a comprehensive solution for building voice-controlled applications, leveraging Python for its versatility and ease of integration with hardware. It offers a robust platform that can be easily adapted and expanded, making it an ideal choice for developers looking to create innovative voice-controlled applications.

Overview: This repository contains the code for the GPT-Driver project, which is a motion planning approach using GPT-3.5 models for autonomous vehicles. It leverages the strong reasoning capabilities of LLMs to generate precise driving trajectories. Key features: - Motion planning using GPT-3.5 models - Representation of planner inputs and outputs as language tokens - Novel prompting-reasoning-finetuning strategy for LLMs Tech stack: Python Why it matters: This project demonstrates the potential of LLMs in solving complex real-world problems like motion planning for autonomous vehicles, showcasing the integration of large language models with autonomous driving technology.

### Overview: This project is an end-to-end automatic speech recognition model implemented using PyTorch, focusing on transformer-based speech recognition. It supports batch parallelization on multi-GPU and is compatible with various datasets for training and evaluation. ### Key features: - **Transformer-based Model**: Utilizes transformer architecture for efficient speech recognition. - **Batch Parallelization**: Supports batch parallelization on multi-GPU for improved performance. - **Multiple Dataset Support**: Compatible with both AI Shell-1 and LibriSpeech datasets for training and evaluation. - **Lightweight and Efficient**: Uses low-rank transformer for lightweight and efficient speech recognition. - **Code-Switched Language Models**: Employs neural-based synthetic data from parallel sentences for code-switched language models. ### Tech stack: PyTorch ### Why it matters: This project is significant for developers interested in speech recognition and natural language processing, as it leverages cutting-edge transformer-based models to achieve high accuracy in speech recognition tasks. The use of PyTorch and low-rank transformers ensures efficient and effective training on multi-GPU setups, making it suitable for large-scale projects and real-world applications.

### Overview: Wn is a Python library designed for exploring information in wordnets, offering multilingual support and interlingual queries via the Collaborative Interlingual Index (Cili). It provides six similarity metrics, functions for taxonomy exploration, and support for lemmatization and Unicode normalization. The library is compatible with the WN-LMF 1.4 format and supports SQL-based backend for fast startup and improved performance. ### Key features: - Multilingual support - Interlingual queries via Cili - Six similarity metrics - Taxonomy exploration - Lemmatization and Unicode normalization - Full support for WN-LMF 1.4 format - SQL-based backend for improved performance ### Tech stack: Python ### Why it matters: Wn is essential for researchers, educators, and developers who need to explore and analyze wordnet data in multiple languages and formats. Its compatibility with various wordnet projects and support for advanced querying make it a valuable tool in fields such as linguistics, language learning, and natural language processing. ### Repository: goodmami/wn ### GitHub stars: 304 ### GitHub description: A modern, interlingual wordnet interface for Python ### Installation: ```sh pip install wn ``` ### Getting Started: ```python import wn en = wn.Wordnet('oewn:2025+') ss = en.synsets('win', pos='v')[0] print(ss.definition()) ``` ### Features: - Multilingual support - Interlingual queries via Cili -

### Overview: DeepSegment is a sentence segmenter that actually works, offering a free API and a self-hostable service. It is designed to segment sentences in various languages, including English, French, and Italian, using pre-trained models and fine-tuning with custom data for optimal performance. ### Key Features: - **Free API and Self-Hosted Service**: Accessible through a free API and a self-hostable service. - **Support for Multiple Languages**: Supports English, French, and Italian. - **Fine-Tuning Option**: Encourages fine-tuning with custom data for better performance. - **API Documentation**: Available at http://bpraneeth.com/docs. ### Tech Stack: Python ### Why It Matters: DeepSegment is a powerful tool for natural language processing tasks, specifically sentence segmentation. Its ability to handle multiple languages and fine-tune models makes it highly versatile and useful for various applications in the field of language processing. ### Repository: GitHub Repository: notAI-tech/deepsegment GitHub Stars: 304 ### Installation: ```bash pip install --upgrade deepsegment ``` ### Supported Languages: - **English**: Trained on various sources. - **French**: Only Tatoeba data. - **Italian**: Only Tatoeba data. ### Usage: ```python from deepsegment import DeepSegment segmenter = DeepSegment('en') segmenter.segment('I am Batman i live in gotham') # ['I am Batman', 'i live in gotham'] ``` ### Using with TensorFlow Serving Docker Image:

### Overview Flappy is a production-ready Language Model (LLM) Application/Agent SDK designed to simplify AI integration in projects. It is an easy-to-use, universally compatible, and production-ready solution that brings the power of AI to developers regardless of their preferred programming language. ### Key Features - **Ease of Use**: Flappy is designed to be as user-friendly as CRUD application development, minimizing the learning curve for developers new to AI. - **Production-Ready**: Beyond research, Flappy is a robust SDK that balances cost-efficiency and sandbox security to provide a stable platform for commercial environments. - **Language-Agnostic**: Flappy integrates seamlessly with any programming language, eliminating the need for Python unless explicitly required by your application. ### Tech Stack Rust ### Why It Matters Flappy is a production-ready Language Model (LLM) Application/Agent SDK designed to simplify AI integration in projects. It is an easy-to-use, universally compatible, and production-ready solution that brings the power of AI to developers regardless of their preferred programming language.

Overview: Semantic Cache is a tool for caching natural text based on semantic similarity, designed to handle synonyms and avoid redundant work in tasks like natural language classification and AI response caching. Key features: - Easily classify natural text into predefined categories - Avoid redundant LLM work by caching AI responses - Reduce API latency by responding to similar queries with already cached values Tech stack: TypeScript Why it matters: Semantic Cache leverages semantic similarity to store cache entries by their meaning, rather than just the literal characters, making it highly effective for tasks that require understanding the context and meaning of text. This approach ensures that similar but not identical pieces of text are treated as semantically equivalent, enhancing the accuracy and relevance of the cache.

Overview: Flama is a production framework designed to serve any model as an API, making it easy to integrate AI models into applications. It supports models from various frameworks and provides a chat UI and native MCP endpoints. Key features: - **Any Framework, One Format**: Package models from scikit-learn, TensorFlow, PyTorch, or LLMs into a single `.flm` format. - **Models on Demand**: Download and package models from the Hugging Face Hub. - **Generative AI Serving**: Serve LLMs with OpenAI-, Anthropic-, and Ollama-compatible endpoints. - **Chatbot Out of the Box**: Every model ships with a polished streaming chat UI at `/chat/`. - **Native MCP**: Expose tools, resources, and prompts to AI agents with a single decorator. - **Rust-Powered Core**: Routing, JSON encoding, request parsing, and compression compiled to native code. - **Production-Ready First**: Go from a packaged model to a running service over the CLI, Python, or container. Tech stack: Python Why it matters: Flama simplifies the integration of AI models into applications by providing a unified API and easy-to-use tools, making it accessible to developers of all skill levels. Repository: vortico/flama Primary language: Python GitHub stars: 304 GitHub description: The production framework for Predictive and Generative AI. Serve any model as an API in one line, with OpenAI/Anthropic/Ollama-compatible endpoints, a built-in chat UI, and

### Overview: 本项目是一个基于Python 3.13、FastAPI、LangChain、LangGraph、React 19、Konva等技术的AI修图智能体项目。它集成了自然语言驱动修图、21种编辑工具、文生图、智能抠图、图层拆分、区域编辑、多步计划等功能，旨在帮助开发者快速掌握AI Agent开发全栈技术。 ### Key features: - **AI 文生图**：输入提示词生成四张候选图，通过SSE实时推送进度。 - **专业级画布编辑器**：支持缩放、平移、图层管理、前后对比滑杆等高级功能。 - **自然语言驱动修图**：通过对话框输入一句话，AI自动规划修图步骤。 - **21种编辑工具**：包括rembg、11种调色工具、裁剪、翻转、缩放、旋转、移动等。 - **智能区域选择**：支持点选、笔刷涂抹模式，计算embedding，响应速度为毫秒级。 - **局部精细编辑**：局部消除和局部替换，AI inpainting和提示词描述替换。 - **图层拆分和独立操作**：图层按语义拆分，独立缩放、调色、移动，互不影响。 - **多步计划和人机协作**：输出结构化JSON计划，拓扑排序确定执行顺序，服务端校验和人机协作。 ### Tech stack: - Python

Overview: HSCRF (Hybrid Semi-Markov CRF) is a neural sequence labeling model that achieves exceptional performance on the CoNLL 2003 NER dataset, achieving an F1 score of 91.38+/-0.10 without additional resources. Key features: - High performance on CoNLL 2003 NER dataset - Uses a hybrid semi-Markov CRF architecture - Achieves state-of-the-art results without corpus or resource augmentation Tech stack: Python, PyTorch Why it matters: HSCRF is a significant advancement in sequence labeling tasks, offering high accuracy with minimal additional resources, making it a valuable tool for researchers and practitioners in natural language processing. Repository: ZhixiuYe/HSCRF-pytorch Primary language: Python GitHub stars: 303

Overview: This project implements a sequence labeling model for Named Entity Recognition using LSTM and CRF, achieving an F1 score of 91.00%. Key features: - LSTM+CRF NER model - Python implementation - High F1 score of 91.00% Tech stack: Python Why it matters: This project demonstrates advanced sequence labeling techniques for Named Entity Recognition, achieving a high F1 score, making it valuable for developers looking to improve their NER models. Repository: ZhixiuYe/NER-pytorch Primary language: Python GitHub stars: 303

### Overview: Mutahunter is an open-source, language-agnostic mutation testing tool powered by language-agnostic LLMs. ### Key features: - Open-source and language-agnostic - Uses language-agnostic LLMs for mutation testing - Supports multiple programming languages ### Tech stack: Python ### Why it matters: Mutahunter provides a comprehensive solution for mutation testing, ensuring robust code quality and reducing the risk of bugs.

### Overview: Memlayer is a lightweight, easy-to-integrate memory layer designed for language models (LLMs). It seamlessly integrates with popular LLMs like OpenAI, Claude, and Gemini, providing persistent, intelligent, and human-like memory capabilities. Memlayer allows agents to recall context, extract structured knowledge, and surface relevant information, all with minimal setup and fast performance. ### Key Features: - **Plug-and-Play**: Install with `pip install memlayer` and get started in minutes. - **Intelligent Memory Filtering**: Three modes (LOCAL/ONLINE/LIGHTWEIGHT) automatically filter important information. - **Hybrid Search**: Combines vector similarity + knowledge graph traversal for accurate retrieval. - **Three Search Tiers**: Fast (<100ms), Balanced (<500ms), Deep (<2s) optimized for different use cases. - **Knowledge Graph**: Automatically extracts entities, relationships, and facts from conversations. - **Proactive Reminders**: Schedule tasks and get automatic reminders when they're due. - **Built-in Observability**: Trace every search operation with detailed performance metrics. - **Flexible Storage**: ChromaDB (vector) + NetworkX (graph) or graph-only mode. - **Production Ready**: Serverless-friendly with fast cold starts using online mode. ### Tech Stack: Python 3.10+ | MIT License | PyPI version ### Why It Matters: Memlayer enhances the capabilities of any LLM, enabling context-aware, intelligent, and human-like interactions. It simplifies the integration process, making it easy to add

Overview: A cross-platform, real-time, offline speech recognition plugin for Unreal Engine, powered by Whisper technology. Key features: - Cross-platform compatibility - Real-time speech recognition - Offline functionality - Open-source Whisper technology Tech stack: C++ Why it matters: This plugin offers developers a reliable and efficient solution for integrating speech recognition into their Unreal Engine projects, leveraging advanced AI technology for improved accuracy and performance.

Overview: Rasa Talk is a chatbot framework built on top of Rasa NLU, designed as an open-source on-premise dialog management system. It aims to provide a simple yet powerful tool for training and deploying chatbots. Key features: - Integrates with Rasa NLU for natural language understanding - Supports multiple third-party chat platforms like Facebook, Telegram, and Slack - Enables training of chatbots with custom intents and expressions - Offers dynamic dialog management with Watson Conversation - Provides easy integration with Rasa NLU for training data generation Tech stack: JavaScript Why it matters: Rasa Talk simplifies the process of building and deploying chatbots by providing a robust framework that can be easily integrated with popular NLU tools, making it accessible to developers without requiring extensive knowledge of machine learning or chatbot technology.