Projects

Overview: This project automates the extraction, structuring, and analysis of financial data from bank statements using advanced techniques like YOLO layout detection, OCR, and LLM-based table extraction. It leverages multiple agent harnesses and local LLMs for efficient processing and reasoning. Key features: - Advanced PDF parsing - Multi-harness agents (CrewAI, Deep Agents, Hermes) - Agent skills for various financial tasks - Local LLMs (LM Studio/Ollama) support - Secure Reasoning Model Support - Full-stack API with FastAPI, PostgreSQL, Celery, and React SPA - GPU acceleration for improved performance Why it matters: This project provides a comprehensive solution for automating the analysis of bank statements, enhancing financial intelligence through robust agent-driven processing and secure reasoning models.

### Overview: Clipper is a Node.js command-line tool designed to extract and convert content from web pages into Markdown format. It leverages Mozilla's Readability library and Turndown to parse HTML and convert it to Markdown, offering a simple and efficient way to save web content for personal use. ### Key Features: - **CLI Interface**: Simplified command-line interface for easy content clipping and conversion. - **HTML to Markdown Conversion**: Converts HTML content into Markdown for easy readability. - **Crawling Support**: Crawls websites and extracts content, converting it to Markdown. - **CLI Usage**: Supports various options for input, output, and format conversion. ### Tech Stack: - TypeScript - Node.js - Readability library - Turndown library ### Why It Matters: Clipper provides a straightforward and efficient method for saving web content for personal use, similar to extensions like Evernote Web Clipper or Notion Web Clipper. It runs entirely in the terminal, eliminating the need for extensions or account sign-ups. ### Repository: philschmid/clipper.js ### GitHub Stars: 630 ### GitHub Description: HTML to Markdown converter and crawler.

Overview: Homer is a Python package designed to enhance the clarity, simplicity, and usefulness of text for readers. It provides insights into an overall text and individual paragraphs, including readability scores, average sentence and word counts, and frequency of "and" words. Key features: - Provides insights into readability, sentence length, and word count. - Identifies and tracks the frequency of "and" words. - Tracks the frequency of "and" words in the text. - Offers paragraph-level statistics, including number and list of compulsive hedgers, intensifiers, and vague words. Tech stack: Python Why it matters: Homer is a valuable tool for writers and content creators looking to improve the readability and clarity of their text. By providing detailed statistics and insights, it helps ensure that the text is easily understandable and engaging for readers. Repository: wyounas/homer Primary language: Python GitHub stars: 630 GitHub description: Homer, a text analyser in Python, can help make your text more clear, simple and useful for your readers. README excerpt: Homer is a Python package that can help make your text more clear, simple and useful for the reader. It provides information on an overall text as well as on individual paragraphs. It gives insights into readability, length of paragraphs, length of sentences, average sentences per paragraph, average words in a sentence, etc. It also tries to identify certain kind of vague words. It also tracks the frequency of "and" words in the text. (More information on all of these follows in the `Acknowledgements` section

### Overview: This repository provides programs to build Retrieval Augmented Generation (RAG) pipelines for Generative AI using LlamaIndex, Deep Lake, and Pinecone. It leverages the power of OpenAI and Hugging Face models to enhance generation and evaluation, making it a valuable resource for developers interested in creating effective generative AI systems. ### Key Features: - Implementing RAG pipelines with LlamaIndex - Using Deep Lake for data storage and retrieval - Integrating Pinecone for efficient data indexing and retrieval - Customizing RAG-driven generative AI systems across various domains ### Tech Stack: - Python - Jupyter Notebook ### Why It Matters: This repository offers practical knowledge on vector stores, chunking, indexing, and ranking, essential for building robust and efficient generative AI systems. By combining text and image data, it provides richer, more informative AI responses, making it a valuable resource for developers looking to enhance their generative AI projects.

Overview: Gyroflow is a project that aims to stabilize video footage using gyroscope data, particularly for drone footage. It leverages Python, OpenCV, and PySide2 to achieve this. Key features: - Video stabilization using gyroscope data - Utility for generating lens undistortion presets - Stabilizing footage using gyro data - Stretching 4:3 video to 16:9 format - Launching utilities with Python and Poetry Tech stack: Python, OpenCV, PySide2 Why it matters: Gyroflow provides a solution for stabilizing drone footage, which is crucial for creating professional-grade videos. It offers a practical way to improve the quality of video stabilization based on gyroscope data, making it a valuable tool for videographers and filmmakers. Repository: ElvinC/gyroflow GitHub stars: 629 GitHub description: [INACTIVE LEGACY VERSION, FIND THE PROJECT HERE: https://github.com/gyroflow/gyroflow]

### Overview: Ravens is a collection of simulated tasks in PyBullet designed for learning vision-based robotic manipulation, focusing on pick and place tasks. ### Key features: - **10 Tabletop Rearrangement Tasks:** Each task has a scripted oracle for imitation learning and reward functions for reinforcement learning. - **Tasks Include:** - **Block Insertion:** Pick up a L-shaped red block and place it into an L-shaped fixture. - **Place Red in Green:** Pick up red blocks and place them into green bowls. - **Towers of Hanoi:** Move disks from one tower to another. - **Align Box Corner:** Pick up a randomly sized box and align one of its corners. - **Stack Block Pyramid:** Stack 6 blocks into a pyramid. - **Palletizing Boxes:** Pick up boxes and stack them on a pallet. - **Assembling Kits:** Arrange objects on a board. - **Packing Boxes:** Pick up boxes and place them tightly into a container. - **Manipulating Rope:** Rearrange a deformable rope. - **Sweeping Piles:** Push piles of small objects into a target goal zone. ### Tech stack: PyBullet ### Why it matters: Ravens provides a comprehensive set of simulated tasks for learning vision-based robotic manipulation, making it an essential resource for researchers and developers in the field of robotic manipulation and reinforcement learning.

### Overview: Fizgig is a comprehensive LoRA & fine-tune studio designed for training, fine-tuning, and repairing models on consumer GPUs. It supports Klein 9B, Flux 3 Image Klein, Krea 2, MiniMax H3, Qwen Image 2.1, MOSS-Image, Z-Image Turbo, SDXL, and Anima. The studio allows users to fix broken LoRAs, explore workbench tools, and train models on photos, video, sound, and voices, from quick LoRAs to the full base model. ### Key features: - **LoRA & LoKR Training**: Fizgig supports training and fine-tuning models using LoRA and LoKR techniques. - **GPU Optimization**: The studio is optimized for consumer GPUs, making it suitable for training on 8 GB GPUs. - **Workbench Tools**: A range of tools for fine-tuning, repairing, and exploring models. - **Model Support**: Supports models like Klein 9B, Flux 3 Image Klein, Krea 2, MiniMax H3, Qwen Image 2.1, MOSS-Image, Z-Image Turbo, SDXL, and Anima. - **Training on Consumer GPUs**: Fizgig allows training on consumer GPUs, making it accessible for users without expensive hardware. - **GPU Readout**: Fixes GPU readout issues on AMD cards. - **Linux Launching Fixes**: Fixes issues with Linux launching. - **Training Speeds**: Improves training speeds, including MOSS-

### Overview: `oh-story-dsh` 是 DeepSeek Harness 的社区插件，提供小说、短剧、游戏和视频解说四个创作工作台。它与 DeepSeek 无隶属关系，通过集成 DeepSeek Harness Web 界面，为用户提供了丰富的创作工具。 ### Key features: - 小说工作台：编辑、审稿、生成封面与去 AI 味。 - 短剧工作台：维护剧本、分镜、视觉设定、视频提示词，生成成片。 - 游戏工作台：快速改编小说为游戏，生成游戏适应性代码。 - 视频解说工作台：预览、质检视频解说项目，流式预览原片、剪后片、成片。 ### Tech stack: Python ### Why it matters: 通过集成 DeepSeek Harness，`oh-story-dsh` 为创作者提供了高效、专业的创作工具，简化了小说、短剧、游戏和视频解说的创作流程，提高了工作效率和作品质量。

Overview: This repository contains fully convolutional Siamese networks for object tracking, enabling high FPS speeds with a slight performance trade-off. It is designed to be used either for tracking alone or for training and tracking. Key features: - Fully Convolutional Siamese Networks for Object Tracking - Supports both tracking only and training and tracking - Pretrained networks available for easy integration into basic trackers - Compatible with ILSVRC15 curation for improved dataset compatibility Tech stack: MATLAB Why it matters: The repository provides a comprehensive solution for object tracking, making it a valuable resource for researchers and developers in the field. The inclusion of pretrained networks simplifies the integration process, while the detailed documentation and step-by-step instructions ensure smooth training and tracking workflows.

### Overview: Word Forms is a Python library that generates all possible forms of an English word, including conjugations, connections between different parts of speech, and pluralization. It accurately converts words like "president" into "presidents", "presidencies", "presidency", and "president". ### Key Features: - **Verb Conjugation**: Converts verbs into their past, present, and future forms. - **Part of Speech Connections**: Connects nouns to adjectives, adverbs, and verbs. - **Pluralization**: Pluralizes singular nouns into their plural forms. - **Adverb Connection**: Connects adverbs to adjectives and verbs. - **Verb Conjugation and Part of Speech Connections**: Accurately generates all possible forms of an English word in one function. ### Tech Stack: - Python ### Why It Matters: Word Forms is essential in Natural Language Processing and Search applications where accurate word treatment is crucial. It provides a robust solution for stemming and lemmatization, ensuring that words like "run" and "ran", "love" and "lovable", and "politician" and "politics" are treated as the same word. This library is invaluable for tasks such as text analysis, machine translation, and information retrieval. ### GitHub Stars: 627 ### Repository: - **Repository**: gutfeeling/word_forms - **Primary Language**: Python - **GitHub Stars**: 627 ### Logo: - **Logo**: ![word forms logo](https://github.com/gutfeeling

### Overview: PULP-Dronet is a deep learning-powered visual navigation engine designed for autonomous navigation of pocket-size quadrotors. It runs on the PULP platform, enabling efficient and lightweight autonomous navigation on nano-sized drones. ### Key Features: - **Autonomous Navigation**: PULP-Dronet enables autonomous navigation of nano-sized drones using deep learning techniques. - **Lightweight Design**: Built on the PULP platform, ensuring ultra-low power consumption and minimal size. - **Versatile Applications**: Suitable for various autonomous drone applications, including visual navigation, obstacle detection, and more. ### Tech Stack: C ### Why it Matters: PULP-Dronet is a groundbreaking solution for autonomous navigation of nano-sized drones, making it a game-changer in the field of autonomous robotics. Its lightweight design and efficient deep learning capabilities make it ideal for applications requiring high performance and low power consumption.

Overview: This repository is a comprehensive AI Agent development tutorial covering various aspects such as LLM agents, RAG, tool use, memory systems, multi-agent systems, LangChain, LangGraph, MCP, and Agentic RL. It provides a structured path from foundational concepts to production engineering. Key features: - 23 chapters across foundations, core capabilities, frameworks, multi-agent systems, production, and capstone projects. - 188 Markdown pages in English and Chinese. - 330+ original SVG diagrams and 5 interactive demos. - Paper-to-practice explanations for ReAct, Reflexion, MemGPT/Letta, GraphRAG, GRPO, MCP, and more. - A reference agent with tools, memory, security gates, evaluation, and a FastAPI service. Tech stack: HTML Why it matters: This repository offers a structured and practical approach to AI Agent development, bridging the gap between basic LLM capabilities and advanced systems like multi-agent and Agentic RL, ensuring a smooth transition from theory to production engineering. Repository: Haozhe-Xing/agent_learning GitHub stars: 619 GitHub description: A systematic AI Agent development tutorial covering LLM agents, RAG, tool use, memory systems, multi-agent systems, LangChain, LangGraph, MCP, and agentic RL.｜从零开始学 AI Agent 开发 | 系统、全面、实战导向的 Agent 开发教程 | 每日自动追踪 arXiv 最新论文 | Learn AI Agent Development from Scratch

Overview: TimeCopilot is an open-source forecasting agent that leverages the power of large language models (LLMs) combined with advanced time series foundation models to automate and explain complex forecasting workflows. It simplifies time series analysis by integrating multiple models and providing easy-to-understand forecasts through natural language queries. Key features: - Unified Forecasting Layer: Combines 30+ time-series foundation models with LLM reasoning for automated model selection and explanation. - Natural-Language Forecasting: Answers questions in plain English, eliminating the need for scripts, pipelines, or dashboards. - One-Line Forecasting: Runs end-to-end forecasts on any dataset in seconds with a single command (`uvx timecopilot forecast <url>`). Tech stack: Python Why it matters: TimeCopilot streamlines time series analysis, making it accessible and professional-grade, by integrating LLMs with state-of-the-art time series models, thus automating and explaining complex forecasting workflows. Repository: TimeCopilot/timecopilot GitHub stars: 626 GitHub description: TimeCopilot: the GenAI Forecasting Agent. Built on LLMs and Time Series Foundation Models, it lets you forecast, cross-validate, and detect anomalies using multiple foundation models through a single API. From finance and energy to web analytics, TimeCopilot turns natural-language queries into production-ready forecasts. README excerpt: <div align="center"> <img src="https://raw.githubusercontent.com/TimeCopilot/timecopilot/main/docs/assets/logos/timecopilot-dark.png#gh-dark-mode-only" alt="

### Overview: BMSG-GAN is a highly stable and robust architecture for generating synchronized multi-scale images, built on the PyTorch implementation of the Multi-Scale Gradient Generative Adversarial Network (MSG-GAN). It addresses the instability during training by allowing gradients to flow from the discriminator to the generator at multiple scales, resulting in stable and high-quality image synthesis. ### Key Features: - **Stable Image Synthesis**: BMSG-GAN provides a stable approach for generating synchronized multi-scale images, addressing the instability issues in GANs. - **PyTorch Implementation**: The code is written in Python using PyTorch, making it easy to integrate with existing projects. - **Training on SageMaker**: The repository includes SageMaker support, making it accessible for training on AWS. ### Tech Stack: Python ### Why It Matters: BMSG-GAN offers a stable and effective method for generating synchronized multi-scale images, making it highly useful in various applications such as image synthesis, style transfer, and more. Its intuitive implementation and robust performance make it a valuable tool for developers looking to improve the stability and efficiency of their GAN-based projects.

Overview: Nanocode is a Rust-based SDK designed for Frontier Embedded Agents, offering a comprehensive set of tools and functionalities to facilitate the development of embedded systems. Key features: - Remote Debugging - Runtime Monitoring - Device Management Tech stack: Rust Why it matters: Nanocode simplifies the development of embedded systems by providing essential tools for remote debugging, runtime monitoring, and device management, making it an indispensable resource for developers working with Frontier Embedded Agents. Repository: gakonst/nanocode Primary language: Rust GitHub stars: 595

### Overview: qwen.cpp is a C++ implementation of Qwen-LM, a large language model designed for real-time chatting. It is based on the ggml library and provides a pure C++ interface, pure C++ tiktoken implementation, and streaming generation with a typewriter effect. The model supports x86, ARM, and NVIDIA GPU hardware, and can be run on Linux and MacOS platforms. It is compatible with Qwen-LM models and offers Python bindings for ease of use. ### Key features: - Pure C++ implementation based on ggml. - Pure C++ tiktoken implementation. - Streaming generation with typewriter effect. - Python binding. ### Tech stack: C++ ### Why it matters: qwen.cpp is a valuable tool for developers looking to integrate Qwen-LM functionality into their projects. It offers a straightforward, efficient way to use Qwen-LM without the need for complex dependencies or external libraries. The pure C++ implementation ensures compatibility and performance, making it a preferred choice for real-time applications. The streaming generation feature provides a seamless typing experience, enhancing user interaction. The Python binding allows for easy integration into existing projects, making it accessible to a wide range of developers. The model's compatibility with newer Qwen models and the ability to be integrated into the llama.cpp project further solidify its importance in the field of language modeling.

Overview: Botonic is an open-source framework designed to build chatbots and conversational apps using React, leveraging Serverless, TensorFlow.js, and popular services like Google Analytics, Contentful, Dialogflow, and Watson. Key features: - Integrates with various messaging apps, websites, and mobile apps. - Supports plugins for analytics, content management, natural language understanding, and more. - Built on top of React, Serverless, and TensorFlow.js. Tech stack: TypeScript, Node.js, npm, and Serverless. Why it matters: Botonic simplifies the creation of conversational apps by abstracting away the complexities of messaging APIs, AI/NLP, and infrastructure management, allowing developers to focus on crafting engaging conversational experiences for users. Repository: hubtype/botonic Primary language: TypeScript GitHub stars: 626

Overview: Fast Sentence Embeddings is a Python library designed to compute sentence embeddings efficiently, making it suitable for large datasets and high-performance computing environments. Key features: - Supports three algorithms for sentence embeddings: unweighted sentence averages, smooth inverse frequency averages, and unsupervised smooth inverse frequency averages. - Provides HUB access to various pre-trained models for convenience. - Full support for Gensim's Word2Vec and FastText classes. - Full support for Gensim's FastText with out-of-vocabulary words. - Induction of word frequencies for pre-trained embeddings. - Incredibly fast Cython core routines. - Dedicated input file formats for easy usage. - Ram-to-disk training for large corpora. - Disk-to-disk training for large corpora. Tech stack: Python Why it matters: Fast Sentence Embeddings is designed to be highly efficient and scalable, making it a valuable tool for large-scale data processing and natural language processing tasks.

Overview: 自媒体内容生产与经营技能集成了模糊需求到最终发布的一整套流程，涵盖创作简报、账号策略、热点与竞品研究、平台原生文案、短视频方案、数字人与动画解说视频制片、多平台视频发布、公众号排版发布、数据复盘和归档，确保内容生产的高效与合规。 Key features: - 模块化的技能集，不绑定特定模型、浏览器、图片、视频、发布或数据服务。 - 平台原生文案，同一主题共享事实与证据，为每个平台分别设计标题、开头、结构和行动。 - 强制确认点，方向、平台、标题、终稿、发布授权五个环节，确保每个环节的准确性。 - 证据优先，关键数字必须有来源，区分事实、判断、推断与建议，不编造数据、体验和收益。 - 视觉体系化，内置 8 套配图风格预设与平台路由，账号偏好优先于预设默认值。 - 数字人制片，头像与声音素材由用户在了解服务商条款后上传，不保存原始素材，只索引生成的交付物路径。 Tech stack: JavaScript Why it matters: 一套通用、模块化的自媒体内容生产与经营技能集，确保内容生产的高效与合规，同时提供灵活的平台原生文案和数字人制片功能，满足不同平台和账号的个性化需求。

Overview: Mapperatorinator is a multi-model framework that generates fully featured osu! beatmaps using spectrogram inputs, aiming to create high-quality, customizable beatmaps. It leverages osuT5 and osu-diffusion for this purpose. Key features: - Automatic generation of osu! beatmaps - Support for all gamemodes - Assist modding capabilities - Colab notebook for local machine usage Tech stack: Python Why it matters: This project aims to automate the creation of high-quality osu! beatmaps from any song, providing a significant boost to the community by reducing the need for manual beatmap creation. Repository: OliBomby/Mapperatorinator GitHub stars: 625 GitHub description: An AI framework for generating and modding osu! beatmaps for all gamemodes from spectrogram inputs.