Projects

### Overview: This repository contains lecture notes and experiments related to deep learning, focusing on various architectures and techniques. It includes slides, teaching guides, and code examples for supervised learning, CNNs, RNNs, Transformers, and more. ### Key Features: - **Supervised Learning**: Slides and teaching guides for understanding supervised learning. - **Datasets and Dataloaders**: Resources for working with datasets and dataloaders. - **Multilayer Perceptron (MLP)**: Code examples for implementing MLPs. - **Convolutional Neural Network (CNN)**: Code examples for CNNs. - **Recurrent Neural Network (RNN)**: Code examples for RNNs. - **Transformer**: Code examples for Transformers. - **Mamba**: A simple implementation of a neural network. - **Optimization and Regularization**: Slides and code for optimization techniques. - **Detection and Segmentation**: Resources for image processing tasks. - **Autoencoder (AE)**: Code examples for autoencoders. - **Variational Autoencoder (VAE)**: Code examples for VAEs. - **Generative Adversarial Network (GAN)**: Code examples for GANs. ### Tech Stack: Jupyter Notebook ### Why it Matters: This repository provides a comprehensive set of resources for learning and experimenting with deep learning concepts, making it an essential tool for developers and researchers in the field.

### Overview: Browserable is an open-source and self-hostable browser automation library designed specifically for AI agents. It enables developers to create browser agents that can navigate websites, interact with forms, click buttons, and extract information. The library is currently at 90.4% on the Web Voyager benchmarks and is highly effective for automating web interactions. ### Key Features: - **Browser Automation**: Navigate websites, fill out forms, click buttons, and extract information. - **Self-Hostable**: Runs on any web server, making it highly accessible and cost-effective. - **Web Voyager Benchmarks**: Achieves 90.4% accuracy, demonstrating its effectiveness in web automation tasks. ### Tech Stack: - JavaScript - Node.js - Docker - Docker Compose ### Why It Matters: Browserable simplifies the process of automating web interactions, making it easier for AI agents to interact with websites and gather information. Its self-hostable nature and high accuracy make it a valuable tool for developers working on AI-driven web applications. ### Repository: - **Repository**: browserable/browserable - **GitHub Stars**: 1208 - **GitHub Description**: Open source and self-hostable browser automation library for AI agents - **GitHub Website**: https://browserable.ai - **Documentation**: https://docs.browserable.ai - **Discord**: https://discord.gg/b6s9fXNjeQ - **Twitter**: https://x.com/browserable ### Quick Start: - **Installation**: Run `npx browserable`

Overview: This repository contains multilingual word vectors for 78 languages, created by aligning them using a linear transformation matrix. These vectors can be used to align the majority of the fastText languages in a single space, without changing any of the monolingual similarity relationships. Key features: - Provides 78 alignment matrices for multilingual word vectors - Aligns the majority of fastText languages in a single space - Can be used for monolingual tasks without changing similarity relationships Tech stack: Python, Jupyter Notebook Why it matters: This repository is useful for researchers and developers who need to compare and align word vectors across multiple languages, enabling better translation and similarity analysis.

### Overview: Instant SECURE Full Stack Apps and AI Agents Turn plain English into a working full-stack app — database, secure API, business logic, and frontend — running on your own hardware, with zero lock-in. An open-source alternative to Lovable, Bolt, and Replit that gives you the whole backend, plus an MCP server that turns every endpoint into a tool for Claude, Cursor, or Codex. Read more about [Hyperlambda](https://hyperlambda.dev). ### Key features: - Secure API - Business logic - Database integration - MCP server - Full-stack app generation - Database-driven AI agents - Secure APIs over legacy databases - Wrappers around third-party APIs - MCP tool servers - Background jobs and automation - Static sites and SPAs ### Tech stack: .NET ### Why it matters: Magic provides a complete solution for developers looking to build secure, scalable, and customizable full-stack applications. By combining the power of Hyperlambda with a robust MCP server, Magic eliminates the need for third-party services, ensuring complete control over your project's architecture and data. This makes Magic an ideal choice for developers who want to build custom, secure applications without the hassle of integrating with external APIs or BaaS providers. Repository: polterguy/magic GitHub stars: 1204 GitHub description: Instant SECURE Full Stack Apps and AI Agents README excerpt: # Instant SECURE Full Stack Apps and AI Agents Turn plain English into a working full-stack app — database, secure

### Overview: BoofCV is a fast, open-source computer vision library written entirely in Java, designed for real-time applications. It includes functionalities such as low-level image processing, camera calibration, feature detection/tracking, structure-from-motion, classification, and recognition. ### Key Features: - **Real-time Processing**: Ideal for applications requiring high-speed image analysis. - **Open Source**: Free and open-source under the Apache License 2.0. - **Java-Based**: Written and maintained in Java, ensuring compatibility and ease of integration. - **Extensive Modules**: Includes modules for low-level image processing, camera calibration, feature detection/tracking, structure-from-motion, classification, and recognition. ### Tech Stack: Java ### Why It Matters: BoofCV is essential for developers looking to implement real-time computer vision solutions in Java applications, providing a robust and efficient framework for tasks such as image processing, camera calibration, and feature detection. ### GitHub Stars: 1,203 ### Repository: The repository is hosted on GitHub at [lessthanoptimal/BoofCV](https://github.com/lessthanoptimal/BoofCV). ### Primary Language: Java ### GitHub Description: BoofCV is a fast computer vision library for SFM, calibration, fiducials, tracking, image processing, and more. ### Build Instructions: To build BoofCV, use the following command: ```bash ./gradlew examplesJar java -jar examples/examples.jar ./gradlew demonstrationsJar java -jar demonstrations/demonstrations.jar ``

### Overview: LightMem is a lightweight and efficient memory management framework designed for Large Language Models and AI Agents, providing a simple yet powerful memory storage, retrieval, and update mechanism. ### Key Features: - **Lightweight & Efficient**: Minimalist design with minimal resource consumption and fast response times. - **Easy to Use**: Simple API design, integrating into your application with just a few lines of code. - **Flexible & Extensible**: Modular architecture supporting custom storage engines and retrieval strategies. - **Broad Compatibility**: Support for cloud APIs (OpenAI, DeepSeek) and local models (Ollama, vLLM, etc.). ### Tech Stack: Python ### Why it Matters: LightMem simplifies the memory management process for Large Language Models and AI Agents, enabling faster development and improved performance.

### Overview: CoCa is a PyTorch implementation of the "CoCa: Contrastive Captioners are Image-Text Foundation Models" paper. This project introduces a novel approach to integrating contrastive learning into conventional encoder-decoder models, achieving state-of-the-art performance on ImageNet with a finetuned encoder. ### Key Features: - **CoCa Implementation**: Implements the CoCa model using PyTorch. - **PaLM Architecture**: Adopted the specific transformer architecture from the PaLM paper for both unimodal and multimodal transformers as well as cross attention blocks. - **State-of-the-Art Performance**: Achieves 91.0% top-1 accuracy on ImageNet with a finetuned encoder. ### Tech Stack: - **Python**: Primary language used for development. - **PyTorch**: Framework used for implementing the model. ### Why It Matters: CoCa provides a unique solution to integrate contrastive learning into conventional image-to-text models, demonstrating significant improvements in accuracy and performance compared to traditional approaches. ### Repository: - **Repository**: lucidrains/CoCa-pytorch - **Stars**: 1201 - **Description**: Implementation of CoCa, a model that uses contrastive learning to enhance conventional image-to-text models, achieving remarkable results on ImageNet. ### Installation: ```bash $ pip install coca-pytorch ``` ### Usage: ```python import torch from vit_pytorch.simple_vit_with_patch_dropout import SimpleViT from vit_pytorch.extractor import Extractor

### Overview: KwaiKEG's KwaiAgents is a comprehensive system designed to leverage Large Language Models (LLMs) for information-seeking and agent-related tasks. This project aims to bridge the gap between cutting-edge AI technology and practical applications, offering a versatile platform for developers and researchers. ### Key features: - **KAgentSys-Lite**: A simplified version of the KAgentSys, featuring a reduced set of tools and slightly lower performance. - **KAgentLMs**: A series of large language models designed to enhance agent capabilities, including planning, reflection, and tool usage. - **KAgentInstruct**: Over 200,000 Agent-related instructions fine-tuned from human input, providing a rich dataset for training and evaluation. - **KAgentBench**: A benchmark dataset consisting of over 3,000 human-edited, automated evaluation scenarios, covering various aspects of agent capabilities. ### Tech stack: Python ### Why it matters: KwaiAgents represents a significant step towards practical application of large language models in agent systems, offering a robust and versatile toolkit for developers and researchers. The project's open-source nature ensures that it can be easily integrated into existing projects and further refined by the community.

### Overview: Lotti is a private logbook with staffed personal AI assistants that syncs end-to-end encrypted across devices. It keeps your memories on your own devices and proposes tasks and goals based on your recorded activities. ### Key features: - **End-to-End Encryption:** All data is synced through a zero-trust relay, ensuring complete privacy. - **Personal AI Assistants:** Agents read your recorded memories and propose tasks and goals. - **Distributed Record:** Memories are stored locally on each device, ensuring no central server holds a readable copy. - **Synced Between Devices:** Data is synced across devices with end-to-end encryption, allowing you to access your memories from any device. ### Tech stack: Flutter and Dart, GPL-3.0, developed since 2016. ### Why it matters: Lotti provides a system of record that is fully owned and controlled by the user, ensuring complete privacy and security. The distributed nature of the record means that losing a device does not compromise your memories, and the end-to-end encryption ensures that no one can read your data without your permission.

### Overview: AutoDL致力于自动化进行任意模态多标签分类，提供一套标准算法流解决复杂分类问题，无需人工干预，只需准备数据即可在短时间内获得性能优异的分类器。 ### Key features: - **全自动**：全流程无需人工干预，自动调节数据、特征、模型和超参。 - **通用性**：支持图像、视频、音频、文本和结构化表格数据，处理任意多标签分类问题。 - **效果出色**：在不同领域获得优异成绩，包括行人识别、行人动作识别、人脸识别、声纹识别、音乐分类、口音分类、语言分类、情感分类、邮件分类、新闻分类、广告优化、推荐系统、搜索引擎、精准营销等。 - **极速/实时**：最快只需十秒即可获得极具竞争力的模型性能，结果实时刷新。 ### Tech stack: Python ### Why it matters: AutoDL为自动化深度学习/机器学习提供了高效解决方案，解决了资源受限、数据倾斜、小数据、特征工程、模型选型、网络结构优化、超参搜索等问题，使得模型的训练和应用变得更加简单和快捷。

### Overview: Proxima is a multi-AI MCP server designed to connect and run various AI models directly within popular coding tools like Cursor, VS Code, Windsurf, and Claude Desktop. It offers a self-healing Python agent, multi-agent delegation, and seamless cross-session memory management, all without requiring any external APIs. ### Key features: - **Multi-AI Support**: Connects and runs multiple AI models including ChatGPT, Claude, Gemini, and Perplexity. - **Free Browser Sessions**: Allows users to run models using free browser login sessions or local offline providers. - **Self-Healing Python Agent**: Ensures continuous performance and reliability. - **Cross-Session Memory**: Maintains session-specific data across multiple runs. - **Routing Modes**: Supports both session routing (default) and BYOK (Bring Your Own Key) routing for more flexible setups. ### Tech stack: JavaScript ### Why it matters: Proxima simplifies the integration of cutting-edge AI models into your development workflow, ensuring that your projects remain up-to-date and efficient. Its seamless integration and robust features make it an indispensable tool for developers looking to leverage the latest AI advancements without the hassle of managing external APIs.

### Overview: SimUtil is a versatile toolset for managing and interacting with Android emulators, iOS simulators, and real devices. It simplifies the process of launching, streamlining, controlling, recording, testing, and automating these devices from a desktop application or command-line interface. ### Key Features: - **Desktop App**: SimUtil Desktop allows you to stream and control multiple devices side by side, including emulators, simulators, and real Android phones and iPhones. It supports recording and automating scenarios using AI agents. - **CLI**: The `simutil` CLI enables you to manage devices, launch and shut down emulators/simulators, connect via Wi-Fi, and interact with Logcat. It supports both desktop and CLI versions. - **Dart Packages**: The project includes Dart libraries for the same device services as the CLI, making it accessible for developers to integrate into their own tooling. ### Tech Stack: - Dart ### Why it Matters: SimUtil streamlines the process of managing and interacting with Android emulators, iOS simulators, and real devices, making it easier to launch, stream, control, record, test, and automate these devices from a desktop app or command-line interface. It simplifies the management of multiple devices and supports AI-driven automation, making it invaluable for developers and testers.

### Overview: Visualize and understand complex neural network architectures with these diagrams. ### Key Features: - YOLO v1 - VGG-16 - Autoencoder - Deep Convolutional Network (DCN) - Recurrent Neural Network (RNN) - Auto Encoder (AE) - Deep Belief Network (DBN) - Restricted BM (RBMs) - ConvLSTM2D for Action Recognition - U-Net - 1D Complex-Valued Neural Network (CVNN) - Feature Pyramid Network (FPN) ### Tech Stack: Diagrams.net ### Why it Matters: These diagrams help developers better understand and visualize the architecture of neural networks, making it easier to debug and optimize models.

### Overview: TrafficGuard/typedai is a TypeScript-first AI platform designed for developers, offering capabilities such as autonomous agents, software developer agents, chatbots, and more. ### Key Features: - **Autonomous Agents**: Capable software engineering agents that have assisted in building the platform. - **Software Developer Agents**: Assist in local repository code editing, task file selection, design creation, and code review. - **AI Chat Interface**: Provides an AI chat interface for developers. - **Slack Chatbot**: Integrates with Slack for additional functionalities. - **Supports Multiple LLM Services**: OpenAI, Anthropic, Gemini, Groq, Fireworks, Together.ai, DeepSeek, Ollama, Cerebras, SambaNova, OpenRouter, X.ai. - **Multi-Agent Extend-Reasoning Implementations**: Configurable human-in-the-loop settings. - **Functional Callable Tools/Integrations**: Extensive integrations with tools like filesystem, Jira, Slack, Perplexity, Google Cloud, Gitlab, GitHub, etc. - **CLI and Web UI Interface**: Provides a powerful command-line interface and a web-based user interface. - **OpenTelemetry Based Observability**: Leverages the extensive Python AI ecosystem through executing Python scripts/packages. - **Run Locally or Deployed on Cloud**: Supports multi-user/SSO with cloud deployment options. - **CLI Usage Guide**: Comprehensive documentation for CLI usage, including `ai` and `aid` scripts. ### Tech Stack: TypeScript ### Why it Matters: TrafficGuard/typedai

Overview: HMTL is a Hierarchical Multi-Task Learning model that combines four semantic tasks: Named Entity Recognition, Entity Mention Detection, Relation Extraction, and Coreference Resolution. It achieves state-of-the-art results on these tasks using PyTorch and AllenNLP. Key features: - State-of-the-art performance on Named Entity Recognition, Entity Mention Detection, and Relation Extraction. - Hierarchical multi-task learning approach. - Pre-trained embeddings available for evaluation with SentEval. Tech stack: Python, PyTorch, AllenNLP. Why it matters: HMTL is a powerful tool for multi-task learning in NLP, offering advanced semantic representation and state-of-the-art performance on challenging tasks.

### Overview: This repository is a curated collection of tools and resources for reverse engineering, binary analysis, and understanding various formats and protocols. It serves as a starting point for those looking to learn about and reverse engineer binaries, applications, firmware, and protocols. ### Key features: - **Tools and Resources**: A diverse set of tools and educational resources for reverse engineering, binary analysis, and understanding file formats and protocols. - **Tool Showcase**: Featured tool REA (Reverse Engineer Anything) by the creators of awesome-reverse-engineering. - **Documentation**: Detailed guides on learning assembly, file identification and triage, native analysis, debugging, dynamic instrumentation, malware analysis, Android and managed code, iOS and macOS, language runtimes and packagers, web and protocols, firmware and hardware, file formats, diffing and analysis frameworks, AI-assisted analysis, practice and communities, related lists, contributing and maintenance, and license information. ### Tech stack: Python ### Why it matters: This repository provides a comprehensive and curated resource for reverse engineering, making it an essential tool for anyone looking to learn about and reverse engineer binaries, applications, firmware, and protocols.

### Overview: VectorDBBench is a comprehensive benchmark tool designed specifically for evaluating vector databases and cloud services. It provides a user-friendly interface for easy benchmarking and cost-effectiveness reports, making it ideal for both professionals and non-professionals alike. ### Key Features: - **Intuitive Interface**: Simplifies the benchmarking process, allowing users to easily initiate and view comparative results. - **Multiple Backends**: Supports multiple vector databases and cloud services, including Milvus, Pinecone, Qdrant, Weaviate, and Elastic. - **Performance Metrics**: Measures key performance metrics such as insertion, vector search, filtered search, and full-text search. - **Registered Datasets**: Offers a wide range of datasets including classic SIFT, GIST, Cohere embeddings, and Hugging Face datasets. - **Cost-Effectiveness Reports**: Provides realistic benchmarking reports for cloud services, helping users make informed decisions. ### Tech Stack: - Python ### Why It Matters: - **Ease of Use**: The intuitive interface simplifies the benchmarking process, making it accessible to users of all technical backgrounds. - **Comprehensive Datasets**: Covers various corpus sizes, dimensions, modalities, distance metrics, and query distributions, ensuring a broad range of benchmark scenarios. - **Realistic Benchmarks**: Provides cost-effectiveness reports for cloud services, offering a more realistic and applicable benchmarking process. - **Community Support**: Sponsored by Zilliz, the leading open-source vectorDB company, providing a reliable and up-to-date benchmarking tool. ### Repository: - Repository:

Overview: This repository contains various machine learning, natural language processing, and deep learning models developed by the author. Key features: - Autoencoder for Audio - Collaborative Filtering - Convolutional Neural Network (Lasagne) - Ensembled Machine Learning - Generative Adversarial Networks - Hyperparameter Tuning Reinforcement Learning - Keras Regularization L2 - Lasagne Neural Networks Regression - Lasagne Neural Nets + Weights - Multinomial Regression - Neural Networks for Regression - NLP + Naive Bayes Classifier - NLP Anger Analysis - NLP Consumer Complaint - NLP Convolutional Neural Network - NLP Doc2Vec - NLP Document Classification Tech stack: Python Why it matters: The author showcases a diverse range of models and techniques in machine learning, natural language processing, and deep learning, providing practical examples for data scientists and developers. Repository: RubensZimbres/Repo-2017 Primary language: Python GitHub stars: 1191 GitHub description: My first Python repo with codes in Machine Learning, NLP and Deep Learning with Keras and Theano

### Overview: coding-agent-search (cass) is a unified, high-performance TUI to index and search your local coding agent session history across 11+ providers (Codex, Claude, Gemini, Cursor, Aider, etc.). It aggregates sessions from various providers into a single, searchable timeline, making it easier to track and analyze your coding sessions. ### Key Features: - **Unified Interface**: A single, intuitive interface to manage and search multiple coding agent sessions. - **Cross-Provider Support**: Supports sessions from Codex, Claude, Gemini, Cursor, Aider, and more. - **Search Capabilities**: Quickly find sessions by keywords, including session details and line numbers. - **CLI and TUI**: Available as a CLI and TUI interface, allowing for easy command-line and graphical access. - **Easy Installation**: Simple installation via script or package manager, ensuring smooth integration into your environment. ### Tech Stack: - **Rust**: The primary programming language used for the backend and frontend components. ### Why It Matters: coding-agent-search (cass) simplifies the process of managing and analyzing your coding agent sessions, making it easier to track performance, identify trends, and troubleshoot issues. It provides a centralized, searchable view of your coding sessions, enhancing productivity and efficiency in your development workflow.

Overview: This project implements a Stock Trading Bot using Deep Q-Learning, focusing on the training and evaluation of a Deep Q-Learning agent on historical stock data. Key features: - Stock Trading Bot - Deep Q-Learning - Model-free Reinforcement Learning - Vanilla DQN, Double DQN, and Dueling Network Architectures Tech stack: Jupyter Notebook Why it matters: This project showcases the application of Deep Q-Learning in stock trading, demonstrating the potential of reinforcement learning techniques in financial markets. Repository: pskrunner14/trading-bot Primary language: Jupyter Notebook GitHub stars: 1190 GitHub description: Stock Trading Bot using Deep Q-Learning README excerpt: # Overview This project implements a Stock Trading Bot, trained using Deep Reinforcement Learning, specifically Deep Q-learning. Implementation is kept simple and as close as possible to the algorithm discussed in the paper, for learning purposes. ## Introduction Generally, Reinforcement Learning is a family of machine learning techniques that allow us to create intelligent agents that learn from the environment by interacting with it, as they learn an optimal policy by trial and error. This is especially useful in many real world tasks where supervised learning might not be the best approach due to various reasons like nature of task itself, lack of appropriate labelled data, etc. The important idea here is that this technique can be applied to any real world task that can be described loosely as a Markovian process. ## Approach This work uses a Model-free Reinforcement Learning technique called Deep Q-Learning (neural