Coding & Development

AI Box is a no-code platform designed for building powerful multimodal AI tools. Users can visually create complex AI workflows using a drag-and-drop interface, chaining together various text, image, and audio models. This platform simplifies the development and deployment of AI applications, making it accessible to individuals without extensive coding knowledge. It aims to streamline the process of integrating different AI capabilities into custom tools, offering instant deployment for developed solutions. AI Box provides a flexible environment for experimenting with and combining diverse AI models to achieve specific functionalities.

MiniGPT-4 is an open-source initiative dedicated to advancing vision-language understanding by integrating advanced large language models. The project offers open-sourced code for both MiniGPT-4 and its successor, MiniGPT-v2, enabling researchers and developers to explore and build upon state-of-the-art vision-language capabilities. It functions as a unified interface, facilitating multi-task learning across various vision and language domains. The project provides detailed instructions for installation, preparation of pretrained LLM weights (including Llama2 Chat and Vicuna), and model checkpoints. Users can launch local demos for both MiniGPT-v2 and MiniGPT-4, with options to optimize GPU memory usage. Training and finetuning details are also provided, making it a comprehensive resource for those working with vision-language models.

Mini-Omni2 is an open-source, omni-interactive AI model designed to provide capabilities similar to GPT-4o, including vision, speech, and duplex interactions. It can understand image, audio, and text inputs, facilitating end-to-end voice conversations with users. A key feature is its real-time voice output and an interruption mechanism during speech, allowing for flexible interaction. The model leverages multimodal modeling by concatenating image, audio, and text features for comprehensive task performance, and uses text-guided delayed parallel output for real-time speech responses. It employs a multi-stage training approach, including encoder adaptation, modal alignment, and multimodal fine-tuning. The model is currently trained on English, though it can understand other languages supported by Whisper for audio encoding, with output remaining in English.

The Marketing Strategy Generator is an open-source project built on the CrewAI framework, designed to automate the creation of detailed marketing strategies. It orchestrates autonomous AI agents to collaborate on complex tasks, from analyzing market trends to developing compelling marketing content. Users can configure environment variables, install dependencies, and customize agent inputs and tasks through YAML files. The tool uses GPT-4o by default, allowing for advanced AI capabilities in generating strategic insights. It provides a structured approach to marketing strategy development, making it a valuable resource for those looking to leverage AI for efficient planning.

Markdowner is a fast and free tool designed to convert any website into LLM-ready markdown data. Built by Supermemory.ai, it addresses the need for structured and predictable data when interacting with Large Language Models, leading to much better AI responses. Key features include LLM filtering to remove unnecessary information, a detailed markdown mode, and an auto-crawler that works without a sitemap. It supports both text and JSON responses and is easy to self-host. The tool utilizes Cloudflare's Browser rendering and Durable objects to spin up browser instances and convert content to markdown using Turndown, offering a robust solution for data preparation.

Melting Pot is an open-source suite of test scenarios specifically designed for multi-agent reinforcement learning (MARL). Developed by Google DeepMind, it offers researchers a robust platform to train and evaluate AI agents in complex social situations. The tool includes over 50 multi-agent games (substrates) and more than 256 unique test scenarios, allowing for the assessment of generalization to novel social interactions like cooperation, competition, and trust. It is built on DeepMind Lab2D and provides tools for interactive play, evaluation of trained models, and example training scripts using frameworks like RLlib. Melting Pot aims to become a standard benchmark for MARL research, with ongoing development to expand its coverage of social interactions and generalization scenarios.

Markdown Validator is an AI-powered tool built on the CrewAI framework, designed to automate the process of reviewing Markdown files for syntax issues. It integrates a custom tool to identify linting errors within Markdown documents. The system then summarizes these errors into a clear list of recommended changes, helping to maintain consistency and quality in documentation. This tool is particularly useful for developers and content creators who frequently work with Markdown and need to ensure their files adhere to established formatting standards. It can be configured to use various models, including locally hosted solutions or the OpenAI API, offering flexibility in deployment. The project also supports agent training, allowing for iterative improvements based on user feedback.

Mergoo is an open-source Python library designed to simplify the process of merging multiple Large Language Model (LLM) experts and then efficiently training the resulting merged LLM. It enables users to integrate knowledge from different generic or domain-specific LLM experts, supporting methods such as Mixture-of-Experts (MoE) and Mixture-of-Adapters (MoA). The library offers flexible merging for each layer and supports popular base models like Llama (including LLaMa3), Mistral, Phi3, and BERT. It is compatible with various trainers including Hugging Face Trainer, SFTrainer, and PEFT, and can run on CPU, MPS, and GPU devices. Mergoo allows for training choices ranging from only the Router of MoE layers to fully fine-tuning the merged LLM.

Mallet is an open-source, Java-based package designed for statistical natural language processing and machine learning applications to text. It provides sophisticated tools for document classification, including efficient text-to-feature conversion, various algorithms like Naïve Bayes and Maximum Entropy, and performance evaluation metrics. Beyond classification, Mallet supports sequence tagging for tasks such as named-entity extraction using algorithms like Hidden Markov Models and Conditional Random Fields. Its topic modeling toolkit offers efficient, sampling-based implementations of Latent Dirichlet Allocation and Hierarchical LDA. The package also includes routines for transforming text documents into numerical representations through a flexible system of "pipes" for tokenizing, stopword removal, and count vector conversion. Mallet is ideal for researchers and practitioners working with large text datasets.

DetGPT is an innovative AI tool designed for object detection through advanced reasoning capabilities. Unlike traditional object detection systems, DetGPT not only identifies objects but also understands complex instructions, allowing it to locate targets based on abstract concepts. For instance, it can identify "blood pressure-reducing foods" in an image by recognizing potassium-rich items like bananas. This ability to provide answers beyond human common sense, such as identifying unfamiliar fruits rich in potassium, makes it a powerful tool for various applications. The project is built upon the open-vocabulary detector GroundingDino and the multimodal conversation model MiniGPT-4, leveraging large language models (LLMs) for its reasoning prowess. It is available as an open-source project on GitHub, providing installation instructions and an online demo for users to explore its features.

DGL-LifeSci is an open-source Python package built on DGL (Deep Graph Library) specifically designed for deep learning applications in life sciences using graph neural networks. It provides a comprehensive suite of tools for researchers and developers, including methods for constructing and featurizing molecular graphs and biological networks, evaluating models, and offering various model architectures. The package also includes training scripts and pre-trained models to accelerate research and development. DGL-LifeSci supports applications such as molecular property prediction and reaction prediction, making it a valuable resource for advancing drug discovery and bioinformatics.

DIG (Dive into Graphs) is a comprehensive open-source library designed for graph deep learning research. Unlike basic graph deep learning libraries, DIG offers a unified testbed for advanced, research-oriented tasks such as graph generation, self-supervised learning on graphs, explainability of Graph Neural Networks, deep learning on 3D graphs, and graph out-of-distribution. It provides unified implementations of data interfaces, common algorithms, and evaluation metrics, allowing researchers to easily implement their own methods and compare them against baseline methods using common datasets and metrics without extensive effort. The library supports various research directions including Graph Augmentation and Fair Graph Learning, and is built on PyTorch Geometric (PyG).

DeepLearningFlappyBird is an open-source project that showcases the application of Deep Reinforcement Learning, specifically Deep Q-learning, to train an AI agent to play the game Flappy Bird. This project is based on the Deep Q Learning algorithm described in "Playing Atari with Deep Reinforcement Learning" and generalizes it to the Flappy Bird environment. It provides a practical, hands-on example for individuals interested in understanding and implementing deep learning algorithms within game environments. The project details the installation process, the architecture of the convolutional neural network used, and the training methodology, including preprocessing steps and hyperparameter annealing. It is an excellent resource for educational purposes and experimentation with AI.

DeepLearningPython is a GitHub repository that offers updated scripts from neuralnetworksanddeeplearning.com, specifically tailored for Python 3.5.2 and integrated with the Theano deep learning library, including CUDA support. This resource provides a practical foundation for individuals looking to learn and implement neural networks. The repository includes three distinct network implementations (network.py, network2.py, network3.py) from the original book, all runnable via a single testing file, test.py. This setup allows users to easily train and evaluate different network configurations, with examples and comments linking back to specific chapters of the book. It's an excellent tool for hands-on learning and experimentation in deep learning.

DeepLearningVideoGames is a project focused on applying deep Q-networks to develop AI agents capable of learning optimal control patterns from visual input in video games. Utilizing reinforcement learning, specifically Q-learning with convolutional neural networks, the system processes raw pixel values from game screens to approximate future expected rewards for actions. The project successfully trained an AI to achieve better than human performance in Pong and is actively working on Tetris. It highlights the potential of deep learning for generalizable high-level control schemes in gaming, demonstrating how AI can learn complex strategies without explicit knowledge of game rules.

deep-learning-v2-pytorch is a comprehensive repository offering projects and exercises for Udacity's Deep Learning Nanodegree program. It features a collection of tutorial notebooks covering diverse deep learning topics, guiding users through the implementation of models such as convolutional networks, recurrent networks, and Generative Adversarial Networks (GANs). The resource also delves into other essential concepts like weight initialization and batch normalization. Beyond tutorials, it provides starting code for Nanodegree projects, which are typically reviewed by Udacity reviewers. This repository is ideal for students and learners looking to gain practical experience and deepen their understanding of deep learning with PyTorch.

deep_learning_curriculum offers an advanced, open-source curriculum designed for individuals seeking to understand the latest developments in deep learning, with a particular emphasis on large language model alignment. It is hosted on GitHub and is intended for those with a strong quantitative background who are already familiar with the fundamentals of deep learning. The curriculum is structured into nine chapters covering topics like Transformers, Scaling Laws, Optimization, Reinforcement Learning, and Alignment. Each chapter includes recommended reading, optional reading, and suggested exercises to facilitate hands-on learning. While challenging, it provides a comprehensive pathway for self-study or mentored learning in this rapidly evolving field.

deep-learning-bitcoin is an open-source project designed to analyze and predict Bitcoin price patterns through deep learning. It employs a unique approach by training models on raw pixel data, mimicking how experienced human traders interpret price charts. The project provides functionalities to download Bitcoin tick data, convert it into 5-minute Open High Low Close (OHLC) representations, and train AI models like AlexNet. Initial results show a 70% accuracy in predicting upward or downward price movements on a dataset of 20,000 samples. The project aims to scale to larger datasets, integrate more complex convolutional neural networks like Google LeNet, and incorporate bar volumes for enhanced prediction accuracy.

IOVISION is an offshore engineering and software development company based in Tunisia, specializing in artificial intelligence and software development services. They offer customized AI-based solutions, web and mobile application development, big data analysis, and software development outsourcing. Their services aim to help businesses integrate AI into operations, digitize processes, and make informed decisions through data analysis. IOVISION also develops its own products, including Finispia (FinTech), MooMe (AI for dairy farm management), and Stile (AI-based personal shopping app). They serve various industries such as Media & Entertainment, Transportation & Logistics, Finance, Agriculture, Healthcare, and Retail/E-commerce, providing multidisciplinary expertise for product development and consulting.

everything-claude-code is a comprehensive performance optimization system designed for AI agent harnesses, originating from an Anthropic hackathon winner. It goes beyond simple configurations, offering a complete system that includes skills, instincts, memory optimization, continuous learning, security scanning, and research-first development. The tool provides production-ready agents, skills, hooks, rules, and configurations that have evolved over months of intensive daily use in building real products. It supports a wide range of AI agent harnesses, including Claude Code, Codex, Cursor, OpenCode, and Gemini, making it a versatile solution for developers working with different platforms. The system emphasizes token optimization, memory persistence, and verification loops to ensure efficient and reliable AI agent operation.

Efficient-3DCNNs offers a PyTorch implementation of "Resource Efficient 3D Convolutional Neural Networks," complete with source code and pretrained models. This repository is designed for developers and researchers working with 3D CNNs, providing a foundation for building and evaluating efficient models. It includes implementations of popular architectures like 3D SqueezeNet, MobileNet, ShuffleNet, and their v2 versions, as well as 3D ResNet and ResNeXt models. The tool supports various datasets such as Kinetics, Jester, and UCF-101, and offers functionalities for training from scratch, resuming training, and fine-tuning with pretrained models. It also includes utilities for data preparation, augmentation, FLOPs calculation, and video accuracy assessment.

fin-ml is an open-source GitHub repository dedicated to providing machine learning and data science blueprints specifically tailored for finance. It features a collection of Jupyter notebooks that contain code for case studies from the O'Reilly book "Machine Learning and Data Science Blueprints for Finance." Users can clone the repository to run these notebooks locally, allowing for hands-on experimentation with the code. The repository covers a wide range of financial applications, including trading strategies, portfolio management, derivatives pricing, asset price prediction, fraud detection, loan default probability prediction, and chatbot development. It also categorizes notebooks by machine learning types such as supervised learning, unsupervised learning, reinforcement learning, and natural language processing, making it a comprehensive resource for developers and data scientists in the fintech domain.

Efficient-Deep-Learning is a comprehensive GitHub repository dedicated to collecting recent methods for deep neural network compression and acceleration. It categorizes techniques into neural architecture re-design or search (NAS), pruning (including structured and unstructured), quantization, matrix/low-rank decomposition, and knowledge distillation (KD). The repository particularly focuses on pruning (with lottery ticket hypothesis or LTH as a sub-topic), KD, and quantization, offering a curated list of relevant papers and surveys dating back to the 1980s. It serves as a valuable resource for researchers and practitioners aiming to improve the efficiency, speed, and compactness of deep learning models.

GRPO-Zero is an open-source project that provides a from-scratch implementation of DeepSeek R1's Group Relative Policy Optimization (GRPO) algorithm. This tool is specifically designed for training large language models using reinforcement learning, emphasizing minimal dependencies and efficient GPU memory usage. It supports training on GPUs with limited VRAM, such as a 24GB RTX 4090, by offloading the optimizer to the CPU, incurring only a small overhead. Key improvements over the original GRPO include token-level policy gradient loss, removal of KL Divergence for reduced GPU memory, and optional overlong episode filtering to stabilize training. The project demonstrates its capabilities by training Qwen2.5 models on a CountDown task, where the model learns to generate mathematical expressions and reasoning.