Projects

### Overview: Prospex is an open-source AI sales and marketing automation tool designed for local businesses. It helps streamline the lead discovery, outreach, and follow-up processes, reducing the need for manual typing and improving efficiency. ### Key Features: - **Lead Discovery**: Finds and enriches business leads based on the provided information. - **Drafting Outreaches**: Writes personalized outreach messages using AI, with the ability to review and approve before sending. - **Workflow Management**: Manages follow-up messages in a single workflow, allowing for easy tracking and approval. ### Tech Stack: - TypeScript ### Why It Matters: Prospex automates the tedious tasks of lead discovery and outreach, allowing businesses to focus on their core sales and marketing efforts. The AI-driven approach ensures that every message is tailored to the specific needs of the business, improving the effectiveness of outreach campaigns.

Overview: Chain-of-Hindsight is a simple next-token-prediction method designed for Reinforcement Learning with Human Feedback (RLHF). It aligns language models with feedback data to improve their performance. Key features: - Supports LLaMA and OPT models - Improves sharding for large models - Adds support for Fully Sharded Data Parallelism (FSDP) - Simplifies data templates for better performance - Enhances support for GPT-J and OPT models - Pre-generates feedback data for training - Simplifies data pipeline Tech stack: Python Why it matters: Chain-of-Hindsight is a practical solution for aligning language models with human feedback, making them more effective in various applications such as language modeling, dialogue systems, and summarization. Repository: haoliuhl/chain-of-hindsight Primary language: Python GitHub stars: 227 GitHub description: Simple next-token-prediction for RLHF

### Overview: Raggo is a lightweight, production-ready RAG (Retrieval Augmented Generation) library written in Go, designed to simplify document processing and context-aware AI interactions. ### Key features: - **Simple RAG**: A basic implementation for straightforward use cases. - **Contextual RAG**: Enhances responses with contextual information from the document. - **Memory Context**: Optimized for chat applications, enhancing conversation flow and memory management. - **Advanced Use Cases**: Supports full processing pipelines, concurrent processing, and rate limiting for robust performance. ### Tech stack: Go ### Why it matters: Raggo simplifies the integration of AI into Go applications, making it easier to build intelligent document processing and context-aware AI solutions without the complexity of deep learning frameworks. ### Repository: teilomillet/raggo ### GitHub stars: 227 ### GitHub description: A lightweight, production-ready RAG (Retrieval Augmented Generation) library in Go. ### README excerpt: # Raggo - Retrieval Augmented Generation Library > A flexible RAG (Retrieval Augmented Generation) library for Go, designed to make document processing and context-aware AI interactions simple and efficient. <p align="center"> <strong>🔍 Smart Document Search • 💬 Context-Aware Responses • 🤖 Intelligent RAG</strong> </p> [](https://pkg.go.dev/github.com/teilomillet/raggo) [![Go Report Card

Overview: This open-source project offers a fully local and private alternative to NotebookLM, allowing users to run their own AI research assistant without relying on external cloud infrastructure. Key features: - Run completely offline - Chat with your documents - Verifiable citations - Local podcast generation - Local audio transcription Tech stack: TypeScript Why it matters: This project provides a self-hosted solution for individuals and companies seeking to leverage AI for internal knowledge management without the need for cloud services. Repository: theaiautomators/insights-lm-local-package Primary language: TypeScript GitHub stars: 227

### Overview: Celeste is a type-safe, modality/provider-agnostic framework for multi-modal AI. It provides a unified interface for interacting with various providers and models, ensuring type safety and ease of use. ### Key features: - **Unified Interface:** One API for OpenAI, Anthropic, Gemini, Mistral, and 14+ others. - **True Multi-Modal:** Supports text, image, audio, video, embeddings, and search. - **Type-Safe by Design:** Full Pydantic validation and IDE autocomplete. - **Zero Lock-In:** Instantly switch providers by changing a single config string. - **Primitives, Not Frameworks:** No agents, no chains, no magic. Just clean I/O. - **Lightweight Architecture:** No vendor SDKs. Pure, fast HTTP. ### Tech stack: Python ### Why it matters: Celeste simplifies the integration and management of multi-modal AI across different providers, making it easier to work with various models and modalities without the complexity of vendor-specific SDKs.

### Overview: This repository is a curated list of awesome topics related to artificial intelligence, covering various aspects such as general AI, learning paradigms, domain-specific applications, and career guidance. ### Key Features: - **General**: Contains resources for general AI topics. - **Compute-specific**: Includes topics related to computing and AI. - **Learning Paradigm**: Focuses on different learning paradigms like adversarial learning, graph learning, reinforcement learning, and supervised learning. - **Domain-specific**: Covers areas like agent-based systems, anomaly detection, audio processing, computer vision, finance, knowledge engineering, mathematics, remote sensing, robotics, software engineering, time series, transformers, and miscellaneous. - **Guideline & Tutorial**: Provides guidelines and tutorials for AI development and career paths. ### Tech Stack: Unknown ### Why It Matters: This repository serves as a comprehensive resource for anyone interested in artificial intelligence, offering a wide range of curated materials and tools across various domains and learning paradigms.

### Overview: This project presents **Diff-Pruning**, an efficient structural pruning method for diffusion models. The method achieves approximately a 50% reduction in FLOPs at a mere 10% to 20% of the original training expenditure, while preserving the generative behavior of the pre-trained models. ### Key Features: - **Efficiency**: Approximately 50% reduction in FLOPs with only a 10% to 20% increase in training cost. - **Consistency**: Pruned diffusion models maintain the generative behavior of the pre-trained models, ensuring consistency with the original models. - **Supports**: Magnitude Pruning, Random Pruning, Taylor Pruning, and Diff-Pruning (a taylor-based method proposed in the paper). ### Tech Stack: Python ### Why it Matters: Diff-Pruning offers a significant improvement in computational efficiency while maintaining the generative capabilities of diffusion models, making it a valuable tool for accelerating training and inference times in diffusion-based models.

Overview: Fine-tune specific SAM model on any task. Key features: - Fine-tune any existing SAM model on a new dataset. - Utilize pre-trained weights to accelerate the training process. - Achieve superior performance on specific tasks without the need for extensive data. Tech stack: Python Why it matters: Fine-tuning allows developers to leverage pre-trained models for high-performance tasks, reducing the computational burden and enabling faster prototyping and deployment.

### Overview: This repository showcases best practices and design patterns for implementing agentic workflows in Python, emphasizing modular, scalable, and reusable components for intelligent automation. It features patterns such as Reflection, Web Access, Semantic Routing, Parallel Delegation, and Dynamic Sharding, facilitating the creation of intelligent, adaptable workflows. ### Key features: - **Reflection**: An iterative content generation and refinement system using an Actor-Critic framework. - **Web Access**: Orchestrates specialized agents for web content retrieval, processing, and summarization. - **Semantic Routing**: Routes user queries to specialized agents based on semantic intent, using a coordinator-delegate architecture. - **Parallel Delegation**: Processes complex queries by delegating entities to specialized agents for parallel processing. - **Dynamic Sharding**: Efficiently processes large datasets by dynamically dividing the workload. ### Tech stack: Python ### Why it matters: This repository provides practical examples and templates for designing agentic workflows, enabling the creation of intelligent, adaptable, and scalable automation solutions.

### Overview: This project is a curated list of Generative AI and Language Model (LLM) APIs designed for developers. These APIs allow developers to integrate advanced AI models into their applications without the need to build them from scratch. The goal is to provide a comprehensive hub for developers to create innovative applications and enhance user experiences. ### Key Features: - **OpenAI**: Offers state-of-the-art generative models for text, image, and speech generation. - **Cohere AI**: Provides a chat API for creating conversational interfaces. - **Anthropic**: An AI safety and research company with powerful models. - **Stability AI**: Offers open-access AI models with minimal resource requirements. - **HuggingFace API**: Access to open-source Generative AI models. - **Groq**: Fastest token generation with language processing units. - **Mistral AI**: Open and portable models for multilingual, code, and advanced reasoning. - **ElevenLabs**: Voice generation API for realistic and natural-sounding voices. - **Perplexity AI**: AI-powered tools for information discovery and summarization. ### Tech Stack: Dart ### Why it Matters: These APIs enable developers to create a wide range of applications, from content creation and virtual assistants to language translation and research assistance. They help developers drive progress in the AI field by providing powerful tools for various industries and tasks. ### Repository: foss42/awesome-generative-ai-apis ### GitHub Stars: 225 ### GitHub Description: A curated list of useful Generative AI & LLM APIs

Overview: 本项目是基于WebnovelWriter改版的OpenCode版本，旨在解决AI写作中的“遗忘”和“幻觉”问题，支持200万字量级的连载创作。 Key features: - 整合三层RAG上下文管理 - 应用故事合约引擎和SSOT事件溯源 - 引入Graph-RAG实体关系图谱和Code Checker硬约束 - 实现一致性错误密度（CED）审查指标 - 提供叙事预报功能 Tech stack: Python Why it matters: 通过将写作知识结构化，确保AI在连载创作中保持初始设定，从而提高创作的连贯性和质量。 Repository: lujih/webnovel-writer-opencode Primary language: Python GitHub stars: 225

Overview: Embedditor is an open-source GUI tool designed to simplify the process of embedding and pre-processing large amounts of text data for vector search, offering a user-friendly interface for editing and optimizing embeddings. Key features: - Join and split chunks of text - Edit embedding metadata and tokens - Exclude specific words or phrases from embedding - Add additional information like URLs or images - Generate HTML-markup for AI search results - Save pre-processed embeddings in .veml or .json formats Tech stack: PHP Why it matters: Embedditor streamlines the embedding process, improving the relevance and efficiency of vector searches while significantly reducing costs associated with embedding and storage.

Overview: Zenith Image Generator is a modern AI image generator that supports multiple providers, including Gitee AI, HuggingFace Spaces, and ModelScope, and is compatible with OpenAI. It offers a sleek dark-mode interface, flexible sizing, secure storage, token rotation, and one-click deployment to Cloudflare Pages. Key features: - Multiple AI providers - Dark mode UI with Gradio-style frosted glass effects - Flexible aspect ratios - Securely stores API keys in localStorage - Token rotation for automatic failover on rate limits - Lightweight history storage - Flow mode for batch generation Tech stack: TypeScript Why it matters: Zenith Image Generator is designed for developers and designers who need a modern and versatile AI image generator, supporting multiple providers and providing a secure, scalable solution for batch generation and deployment. Repository: WuMingDao/zenith-image-generator GitHub stars: 225

### Overview: This repository is an official PyTorch implementation of the paper "Unsupervised Representation Learning from Pre-trained Diffusion Probabilistic Models" (NeurIPS 2022). It provides a comprehensive set of tools and scripts for training, evaluating, and sampling from the PDAE model. The repository includes datasets, scripts for training different components of the model, and evaluation metrics to assess performance. ### Key features: - **Training Scripts**: Scripts for training Diffusion Probabilistic Models (DPMs), representation learning, and manipulation tasks. - **Datasets**: Pre-trained datasets for diffusion models, including LMDB-ready datasets for various image and text-based tasks. - **Evaluation Metrics**: SOTA performance metrics for autoencoding, denoising, posterior mean gap measure, and manipulation tasks. - **Example Scripts**: Sample scripts for autoencoding, denoising, posterior mean gap measure, interpolation, and manipulation tasks, demonstrating how to modify and run the scripts. ### Tech stack: Python ### Why it matters: This repository is essential for researchers and developers working on unsupervised representation learning and diffusion models. It provides a robust framework for training and evaluating models that can be applied to various image and text-based tasks, making it a valuable resource for researchers and practitioners in the field of machine learning and computer vision.

### Overview: Opik MCP Server is a Model Context Protocol (MCP) server designed for the open-source LLM observability and evaluation platform, Opik, built by Comet. It allows users to read traces, log scores, and manage prompts directly from their AI hosts such as Claude Code, Cursor, VS Code Copilot, or Codex. ### Key features: - **Read Traces**: Retrieves and displays traces from Opik projects. - **Log Scores**: Records and logs scores for specific traces. - **Manage Prompts**: Saves and manages prompt versions for easy reference. ### Tech stack: Python ### Why it matters: Opik MCP Server simplifies the integration of Opik with AI hosts, enabling LLM engineers to leverage their existing Opik setup and AI assistants like Claude Code, Cursor, VS Code Copilot, or Codex. This tool streamlines the process of monitoring and evaluating language models within their development workflows, enhancing observability and feedback loops.

Overview: Gemini PDF Chatbot is a Streamlit-based application that allows users to engage in natural language conversations with a conversational AI model trained on PDF documents. It extracts text from uploaded PDF files and provides detailed responses based on the document context. Key features: - Upload multiple PDF files - Extract text from uploaded PDF files - Use the Gemini conversational AI model to answer user questions - Provide a chat interface for interaction Tech stack: Python Why it matters: This application enhances the interaction with PDF documents by leveraging advanced AI models, making it easier to extract and understand information from various documents.

### Overview: This repository is a curated collection of papers and resources focused on anomaly detection foundation models. It includes models and techniques that leverage large language models, large vision-language models, graph foundation models, time series foundation models, and more. The repository aims to provide a comprehensive overview of the latest advancements in anomaly detection, making it a valuable resource for developers and researchers. ### Key Features: - **Large Language Models (LLM):** Models like AnomalyGPT and AnomalyGFM. - **Large Vision-Language Models (VLM):** Models like WinCLIP and AnomalyCLIP. - **Graph Foundation Models:** Models like CLIP-AD and AnomalyGFM. - **Time Series Foundation Models:** Models like AnomalyGFM and AnomalyGFM. - **Zero-Shot and Few-Shot Anomaly Detection:** - WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation. - CLIP-AD: A Language-Guided Staged Dual-Path Model. - AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models. - AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection. - AnomalyCLIP: AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection. - AnomalyCLIP: AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection. - AnomalyCLIP: AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection.

Overview: This project presents an innovative method for text-driven image editing, enabling precise image manipulation without the need for user-provided masks or sketches. It utilizes an existing pre-trained text-to-image model to find and edit regions aligned with textual prompts, achieving high fidelity and realism in image editing tasks. Key features: - Text-driven image editing - Flexible region-based editing - Compatible with current image generation models - Handles complex prompts with multiple objects, sentences, or paragraphs Tech stack: Python Why it matters: This method offers a new approach to image editing that simplifies the process by leveraging language as a natural interface, making it accessible and effective for a wide range of applications. Repository: yuanze-lin/Learnable_Regions Primary language: Python GitHub stars: 225 GitHub description: [CVPR 2024] Official code for "Text-Driven Image Editing via Learnable Regions"

Overview: Ablation is an AI-assisted reverse engineering (AIARE) framework designed to automate the analysis of binary code and machine instructions. Key features: - Native CPUs support for Alpha AXP, ARC, ARM, Intel MCS, LoongArch, MIPS, PowerPC, RISC-V, S/390, x86, and Xilinx. - Automotive ECUs support for Infineon, Infineon C166, Motorola, Renesas SH, TI DSP, and V850 / RH850. - VMs & Runtimes support for ARK, BEAM, DEX, Java, MicroPython, and Python. Tech stack: Python Why it matters: Ablation simplifies the process of reverse engineering by leveraging AI to automate the analysis of binary code and machine instructions, making it easier and more efficient for developers to understand and modify software. Repository: Ablation-Tool/ablation Primary language: Python GitHub stars: 224 GitHub description: Ablation is an AI-assisted reverse engineering (AIARE) framework. README excerpt:  [](docs/standards-ablation.md) [![CISA Secure by Design](https://img

Overview: CopywriterProAI is the world's first open-source AI content writing platform designed to help users create SEO-friendly blog posts, ad copy, and more. Powered by cutting-edge AI technology, it streamlines the writing process, ensuring high-quality output without the need for manual labor. Key features: - Seamless integration with popular content management systems (CMS) - Advanced AI-driven SEO optimization - Customizable packages for different levels of expertise - Real-time plagiarism detection and plagiarism checker - User-friendly interface for easy content creation Tech stack: The backend of CopywriterProAI is built using Node.js, Yarn, MongoDB, Docker, and Docker Compose. It leverages the power of AI to automate content creation, making it accessible to users of all skill levels. Why it matters: CopywriterProAI empowers users to create high-quality content efficiently, saving time and effort while ensuring optimal SEO performance. Its open-source nature means it can be easily integrated into existing workflows and improved upon by the community, making it a valuable tool for content creators and marketers alike.