Projects

Overview: This repository contains AI-powered software development patterns that integrate generative AI into various stages of the software development lifecycle using Amazon Q Developer, Amazon Q Business, and Amazon Bedrock. Key features: - Demonstrates practical approaches for leveraging AWS's generative AI capabilities - Contains patterns for different SDLC phases - Integrates Amazon Q Developer, Amazon Q Business, and Amazon Bedrock Tech stack: Python Why it matters: This collection of patterns showcases how to effectively integrate AI into software development, improving productivity, quality, and delivery times. Repository: aws-samples/sample-ai-powered-sdlc-patterns-with-aws GitHub stars: 90 GitHub description: This repo contains AI-powered software development patterns showing how to integrate generative AI in different stages of software development lifecycle using Amazon Q Developer, Amazon Q Business and Amazon Bedrock. This collection of patterns demonstrates practical approaches for leveraging AWS's generative AI capabilities across the software development lifecycle (SDLC). The patterns are designed to help development teams enhance productivity, improve quality, and accelerate delivery through AI-powered development. Important: this patterns use various AWS services and there are costs associated with these services - please see the [AWS Pricing page](https://aws.amazon.com/pricing/) for details. You are responsible for any AWS costs incurred. No warranty is implied in this example. Deployment Instructions: 1. Create an AWS account if you do not already have one and login. 2. Install Git on your local machine. 3. Create a new directory and navigate to that directory in a terminal. 4. Clone this

### Overview: This repository contains code for the book "Python Deep Learning Projects," which aims to demystify deep learning models for various applications such as natural language processing (NLP), computer vision, and audio recognition using TensorFlow and Keras. ### Key features: - Code for various deep learning projects - Focus on practical application of deep learning models - Covers NLP, CV, and ASR - Uses TensorFlow and Keras frameworks - Aimed at intermediate machine learning engineers and software developers ### Tech stack: Jupyter Notebook ### Why it matters: This repository serves as a practical guide for intermediate machine learning engineers, software developers, technology architects, and programmers interested in deep learning, particularly applied deep learning. It provides a comprehensive look at the data science pipeline, from model building to training and validation, using tools like TensorFlow, Keras, and Google Cloud.

### Overview: AudioToken is an innovative project that leverages text-conditioned diffusion models to generate high-quality audio-to-image (A2I) results. This approach uses a pre-trained audio encoding model to adapt latent diffusion models, enabling the generation of images conditioned on audio recordings. ### Key Features: - **Text-Conditioned Diffusion Models**: Utilizes latent diffusion models trained for text-to-image generation to generate images conditioned on audio recordings. - **Latent Representation**: Encodes audio into a new token that acts as an adaptation layer between the audio and text representations, making the model lightweight. - **Pre-trained Model**: Uses a pre-trained audio encoding model to pre-train and fine-tune the model, improving performance. - **Objective and Subjective Metrics**: Superior performance compared to baseline methods in both objective and subjective metrics. ### Tech Stack: - **Programming Languages**: Python - **Frameworks**: PyTorch ### Why It Matters: - **Performance Improvement**: The proposed method outperforms existing baselines in terms of both objective and subjective metrics. - **Lightweight Optimization**: The model requires a small number of trainable parameters, making it suitable for lightweight optimization. - **Versatility**: The approach can be applied to various modalities, enhancing the versatility of diffusion models. ### Repository: - **Repository**: [GitHub Repository](https://github.com/guyyariv/AudioToken) - **Stars**: 90 - **Description**: The official PyTorch implementation of "AudioToken: Adaptation of Text-Conditioned Diffusion Models for Audio-to-

### Overview: Pluto is a comprehensive platform that enables the deployment of AI agents within Salesforce, utilizing standard licenses and any model provider, including OpenAI, Azure OpenAI, or your own. It simplifies the process with a dedicated engine, Pluto Core, and optional packages like Pluto Model Adapters and Pluto Console for enhanced functionalities. ### Key Features: - **Pluto Core**: The engine that runs, manages, and approves AI agents, ensuring data protection and recovery. - **Pluto Model Adapters**: Ready-made connections to popular model providers, making integration straightforward. - **Pluto Console**: An admin console for setting up agents, monitoring runs, and approving changes. ### Tech Stack: Apex ### Why it Matters: Pluto simplifies the integration of AI into Salesforce, offering robust features like model calls, tool approvals, and data protection, all while maintaining standard Salesforce licenses. The platform's flexibility and ease of use make it an ideal solution for businesses looking to leverage AI without the complexity of custom development. ### Repository: iamsonal/pluto ### GitHub Stars: 89 ### GitHub Description: Run AI agents inside Salesforce on standard licenses, with any model provider. Pluto Core handles model calls, tools, approvals, recovery, and a record of every run. Works with OpenAI, Azure OpenAI, or your own model provider.

### Overview: Intuitive Robots' "Beso" project introduces a novel approach to goal-conditioned imitation learning using score-based diffusion policies. This method aims to enhance the accuracy and efficiency of learning from demonstrations by conditioning the imitation process on specific goals. ### Key Features: - **Goal Conditioned Imitation Learning**: The project focuses on learning from demonstrations while explicitly conditioning the imitation process on specific goals, improving the quality of learned policies. - **Score-Based Diffusion Policies**: Utilizes score-based diffusion models to generate realistic trajectories from demonstrations, ensuring the learned policies are more robust and effective. - **Hydra Configuration Manager**: All configurations are managed using Hydra, a hierarchical configuration manager, making it easier to manage and update settings across different environments. ### Tech Stack: - **Python**: The primary programming language used for the project. - **WandB**: Used for logging and monitoring all runs, enabling easy tracking and analysis of training processes. ### Why It Matters: - **Enhanced Learning Quality**: By conditioning the imitation process on specific goals, the project aims to produce more accurate and efficient learned policies. - **Robust Trajectory Generation**: Score-based diffusion models generate realistic trajectories from demonstrations, ensuring the learned policies are more reliable and effective. - **Versatile Environment Support**: The project supports both the Franka Kitchen and the Block Push environments, providing a flexible framework for imitation learning applications. ### Repository: - **GitHub Repository**: intuitive-robots/beso - **GitHub Stars**: 90 - **GitHub Description**: Official code for "Goal Conditioned Imitation Learning

Overview: Backtalk is a Claude Code agent that allows you to talk to your Claude Code agent out loud, providing real-time responses through your speakers. It uses local models and no API keys, making it accessible and cost-effective. Key features: - Hold a key, talk, release. Your words are transcribed locally and handed to a live Claude Code session. - It's YOUR agent talking. The session runs in the folder whose CLAUDE.md defines your assistant. - Interrupt it. Press the key while it's talking and it shuts up and listens. - It asks before it acts, in plain words. - The voice console allows session control by voice. - It can pick up where it left off and thinks out loud. Tech stack: Python Why it matters: Backtalk is a practical tool for anyone looking to integrate Claude Code into their workflow, providing a hands-free and cost-effective way to interact with Claude Code. Repository: jaredrhod/backtalk Primary language: Python GitHub stars: 89

Overview: Str2Str is a score-based framework designed for zero-shot protein conformation sampling, utilizing a neural score network trained on the PDB database to generate diverse conformations of a given protein structure. Key features: - Implements a zero-shot approach for protein conformation sampling. - Utilizes a neural score network for sampling based on a score network trained on the PDB database. - Supports both diffusion/flow matching architectures. - Accommodates input protein structures obtained from folding modules like AlphaFold2 or ESMFold. Tech stack: Python Why it matters: Str2Str provides a novel and effective method for generating diverse protein conformations using a score-based framework, offering a zero-shot solution for protein structure sampling.

### Overview: Tiled Diffusion is an innovative approach to generating seamless tileable images using diffusion models. This method extends the capabilities of diffusion models to support the creation of cohesive tiling patterns across various domains of image synthesis, enabling seamless integration of multiple images. ### Key Features: - **Seamless Tiling:** Automatically generates cohesive tiling patterns across diverse domains of image synthesis. - **Many-to-Many Connections:** Supports complex many-to-many connections, enabling seamless integration of multiple images. - **Automatic Tiling Process:** Eliminates the need for manual intervention, enhancing creativity in various applications. ### Tech Stack: - Python ### Why It Matters: Tiled Diffusion is a significant advancement in the field of image synthesis, providing a novel solution for generating seamless and cohesive tileable images. This method addresses the limitations of traditional tile creation by automating the process, making it scalable and flexible, and opening up new possibilities in applications such as texture creation, video game asset development, and digital art.

### Overview: The Meshy MCP Server is a powerful tool for Meshy AI, enabling the creation, management, and download of 3D models, textures, images, rigged characters, and animations through natural conversation. ### Key Features: - **28 Tools**: Covering the full Meshy API, including 3D Generation, Creative Lab, Post-Processing, Rigging & Animation, Conversion, Image Generation, Task Management, Workspace, 3D Printing, and Account Management. - **Safe Retries**: A create request is never resent after a timeout or server error, ensuring tasks are never billed twice. - **Text to 3D**: Generates 3D models from text descriptions, with a preview and refine pipeline. - **Image to 3D**: Converts single or multiple images into 3D models. - **Meshy 7.1**: Supports advanced features like Smart Topology, 8K Textures, and Multi-view Retexture. - **Meshy 6 Lite**: Offers fast, 5-credit mesh generation for game assets. - **Auto-Rigging & Animation**: Adds skeletons and animations to characters. - **Creative Lab**: A single tool for photo to product, with 9 products and configurable build options. - **Format & Size Utilities**: Tools for converting formats, resizing images, and unwrapping UVs. - **2D Image Models**: Supports text-to-image and image-to-image conversion. ### Tech Stack: TypeScript ### Why It Matters: The Meshy MCP Server simplifies the

### Overview: A free, remote, and easily installable English practice tool for Turkish speakers. It includes speech and writing practice using AI, A1-C1 language lessons, and grammar exercises. It supports Windows and Linux, and can be installed on Apple Silicon Macs as well. It uses Whisper, Gemma 4, Kokoro, and EMA Lightning for speech recognition and pronunciation correction. ### Key features: - Speech and writing practice using AI - Language lessons from A1 to C1 level - Grammar exercises - Pronunciation correction - Reading and writing exercises - Vocabulary lists and quizzes - Daily writing and reading habits - Speed tests using YDS format - Progress tracking - Remote installation and use ### Tech stack: Python ### Why it matters: This tool is designed for Turkish speakers who want to improve their English skills, especially those who work. It provides a practical way to practice English through speech and writing exercises, language lessons, and pronunciation correction, making it useful for both beginners and advanced learners.

### Overview: PoloX AI is an open-source platform designed for multimodal AI generation, creation, and editing. It leverages agent conversations and an infinite canvas to unify these processes, allowing seamless interaction and continuous workflow. ### Key Features: - **Agent Conversations**: Engaging agent interactions that guide and refine the creation process. - **Infinite Canvas**: A continuous workspace where all interactions and creations are visible and editable. - **Multimodal Support**: Integrates various forms of media, including images, videos, and text, for comprehensive content creation. - **Real-time Updates**: Real-time adjustments and feedback from the AI, ensuring the project stays current and accurate. ### Tech Stack: - TypeScript - Nuxt, Vue, SQLite - WaveSpeed API for AI inference ### Why It Matters: PoloX AI provides a unified workflow for multimodal AI creation, enabling creators to generate, edit, and refine content effortlessly through agent-native interactions. The platform's seamless integration and real-time updates make it an indispensable tool for modern creatives. ### GitHub Stars: 91 ### Repository: - Repository: saihhold-zhao/polox_ai - Primary Language: TypeScript - GitHub Description: An open-source, agent-native platform for multimodal AI generation, built on DeepSeek Harness. Every interaction lives in agent conversations and an infinite canvas, unifying creation, generation, and editing. - GitHub Stars: 91

Overview: VideoRAG is a Retrieval-Augmented Generation (RAG) framework designed to enhance the accuracy of generative AI models by integrating video retrieval with textual information. Key features: - Dynamic Video Retrieval - Multimodal Representation - Efficient Frame Selection Mechanism - Textual Information Extraction - State-of-the-Art Performance Tech stack: Python Why it matters: VideoRAG leverages the rich context provided by videos to improve the accuracy of generative AI models, surpassing existing baselines on various benchmarks.

### Overview: DermoSegDiff is a novel boundary-aware segmentation model designed for skin lesion delineation, showcasing exceptional performance on various dermatological datasets. ### Key Features: - **Boundary-aware Segmentation**: Incorporates boundary information during training to improve segmentation accuracy. - **Loss Function**: Introduces a novel loss function prioritizing boundaries, reducing the significance of other regions. - **U-Net-based Denoising Network**: Integrates noise and semantic information effectively within the network. - **Performance Superiority**: Demonstrates significant superiority over existing CNN, transformer, and diffusion-based approaches. ### Tech Stack: - **Python v3.7 or higher** - **Pytorch v1.7 or higher** - **CUDA 11.1 or higher** - **Hardware Requirements**: 12GB or larger GPU memory (with low memory, adjust parameters accordingly) ### Why It Matters: DermoSegDiff offers a more accurate and effective method for skin lesion segmentation, aiding in early detection and diagnosis of dermatological conditions. ### Repository: - **Repository**: xmindflow/DermoSegDiff - **GitHub Stars**: 89 - **GitHub Description**: [MICCAI 2023] DermoSegDiff: A Boundary-aware Segmentation Diffusion Model for Skin Lesion Delineation - **Primary Language**: Python - **Download Links**: ISIC2018: [[Download](https://uniregensburg-my.sharepoint.com/:f:/g/personal/say26747_ads_uni-regensburg_de/E

### Overview: LipVoicer is an official implementation of a lip-to-speech method that generates speech from silent videos using lip reading and a diffusion model. It is based on the ICLR paper "LipVoicer: Generating Speech from Silent Videos Guided by Lip Reading" and is designed to perform well on in-the-wild datasets with hundreds of unique speakers and an unrestricted vocabulary. ### Key features: - Generates speech from silent videos using lip reading and a diffusion model. - Performs well on LRS2 and LRS3 datasets. - Pretrained models available for LipVoicer. - Benchmarks provided for test audio signals generated by LipVoicer. - Pretrained checkpoints available for LipVoicer. ### Tech stack: Python ### Why it matters: LipVoicer offers a practical and effective solution for generating speech from silent videos, making it valuable for applications that require lip-reading capabilities. Its performance on real-world datasets highlights its utility in scenarios where lip-reading is challenging or impossible.

### Overview: GPTalk is a Python-based GPT-3 client designed for Windows and Unix systems. It supports text and speech functionalities in multiple languages, making it versatile for various applications. The client is designed to be user-friendly and integrates seamlessly with Discord, Twitter, and a donation link. ### Key Features: - **Text and Speech Support**: GPTalk allows users to interact with GPT-3 using text or speech, supporting multiple languages. - **Free Text2Image**: The client includes a free text-to-image feature using DALL-E, making it useful for creative tasks. - **Language Support**: Supports over 100 languages for both text and speech modes. - **Integration**: Integrates with Discord, Twitter, and a donation link, making it accessible to a broader audience. ### Tech Stack: - **Python**: The primary programming language used for development. - **DALL-E**: Integrated for text and speech generation, providing visual outputs. ### Why It Matters: GPTalk simplifies the interaction with GPT-3, making it accessible and useful for a wide range of applications, including creative writing, language translation, and more. Its integration with popular platforms like Discord and Twitter ensures broad reach and community engagement. ### Repository: - **Repository**: 0ut0flin3/Talk2GPT - **GitHub Stars**: 89 - **GitHub Description**: GPT-3 client for Windows and Unix with memories management that supports both text and speech in any language. Includes a free text2image feature. ### Requirements: -

### Overview: This repository compiles the most comprehensive and up-to-date comparison of AI image generation models, providing developers with the necessary information to choose the right model and API for their needs. It includes real per-image pricing and detailed insights into the models' capabilities and performance. ### Key Features: - **Model Comparison**: Lists every AI image generation model available, including their APIs and pricing. - **API Integration**: Maps each model to the corresponding APIs that serve it, ensuring developers can easily integrate them into their projects. - **Price Transparency**: Provides real per-image pricing, including retail API rates verified in August 2026, and notes that prices can change rapidly, so it's essential to confirm against the provider. - **Quality and Performance**: Offers a detailed analysis of each model's quality, performance, and what it excels at, helping developers make informed decisions. ### Tech Stack: - **GitHub Stars**: 88 ### Why It Matters: This repository is crucial for developers looking to generate high-quality images efficiently. It simplifies the process of selecting the right model and API, ensuring that developers can focus on their core work without worrying about technical details.

Overview: This repository is an official implementation of "Vision-by-Language for Training-Free Compositional Image Retrieval" (CIReVL) at the 2024 ICLR conference. CIReVL aims to achieve zero-shot compositional image retrieval (ZS-CIR) without requiring training task-specific models, leveraging large-scale vision-language models (VLMs) and large language models (LLMs) for efficient and scalable retrieval. Key features: - CIReVL: A training-free compositional image retrieval pipeline combining large-scale VLMs and LLMs. - ZS-CIR: Zero-shot compositional image retrieval, achieving competitive performance on benchmark datasets. - Modular language reasoning: Reconstructing captions using a pre-trained generative VLM and a LLM. - Human-understandable: Composing images and text in a modular fashion, making it scalable and post-hoc re-alignable. Tech stack: Python, PyTorch, OpenCV, OpenCLIP, transformers, salesforce-lavis, CLIP. Why it matters: CIReVL offers a practical solution for zero-shot compositional image retrieval, demonstrating significant improvements over supervised methods and enabling easy scalability without re-training, making it accessible for various applications and datasets. Repository: ExplainableML/Vision_by_Language Primary language: Python GitHub stars: 89 GitHub description: [ICLR 2024] Official repository for "Vision-by-Language for Training-Free Compositional Image Retrieval"

Overview: This repository provides a step-by-step implementation of a Text2Video model using PyTorch, focusing on a 3D U-Net architecture with attention mechanisms. The model is trained on a dataset of 10,000 training steps and demonstrates results on various prompts. Key features: - Text2Video model implementation - 3D U-Net architecture with attention mechanisms - Training on a 10,000-step dataset - Results on prompts such as "A person holding camera" and "Spaceship crossing the bridge" Tech stack: Python Why it matters: This project offers a practical guide for developers interested in creating video generation models using text-to-image techniques, showcasing the implementation of a complex model architecture and training process. Repository: FareedKhan-dev/text2video-from-scratch GitHub stars: 89

### Overview: RecDiff: Diffusion Model for Social Recommendation RecDiff is a novel framework that leverages the power of diffusion models to address the challenge of noisy social connections in social recommendation systems. By integrating hidden-space diffusion processes with graph neural networks, RecDiff proposes a method to surgically remove noise from social signals, leading to improved performance on various benchmarks. ### Key Features: 1. **Multi-Step Social Denoising**: RecDiff uses forward-reverse diffusion to progressively remove noise from social signals. 2. **Task-Aware Optimization**: The framework is designed to be task-oriented, ensuring that downstream tasks are optimized during the diffusion process. 3. **Hidden-Space Processing**: Efficient diffusion in a compressed representation space is used to handle the noise effectively. 4. **Adaptive Noise Handling**: The framework dynamically adapts to varying levels of social noise, ensuring optimal performance. ### Tech Stack: Python RecDiff is implemented using Python, leveraging libraries such as PyTorch for deep learning and TensorFlow for graph neural networks. The project is built on the latest Python 3.8 and PyTorch 1.12, ensuring compatibility and performance. ### Why It Matters: SOTA Performance on 3 Benchmarks RecDiff has achieved state-of-the-art performance on three key benchmarks in social recommendation systems. By addressing the noise contamination issue, RecDiff significantly improves recommendation accuracy and user satisfaction, making it a game-changer in the field. ### Repository: HKUDS/RecDiff The repository is hosted on GitHub, where it has been starred by 89

Overview: ML Guide is a comprehensive repository designed for machine learning enthusiasts and professionals alike, offering a wide range of resources, tutorials, projects, and datasets to enhance learning and skill development. Key features: - A variety of projects and code samples for practice and real-world experience. - Open-source, allowing anyone to contribute and modify the content. Tech stack: Jupyter Notebook Why it matters: ML Guide serves as a valuable resource for anyone interested in machine learning, providing a structured and collaborative environment for learning and skill enhancement.