Projects

Overview: DeepMem is a fast, cost-effective AI memory layer that seamlessly integrates with Mem0 API, allowing for a 2x faster response and 10x lower cost. It can be self-hosted for free or managed on the cloud, making it accessible without any code changes or authentication. Key features: - Self-hostable - Fully compatible with Mem0 API - 2x faster retrieval - 10x lower cost - 5-minute migration process - No auth, no payment, no lock-in - Cloud version available for managed deployment Tech stack: Python Why it matters: DeepMem offers a fast, cost-effective solution for AI memory layer integration, making it ideal for developers looking to improve their applications' performance and cost-effectiveness. Repository: deepmemteam/deepmem GitHub stars: 114 GitHub description: Drop-in AI memory layer with 2x faster retrieval and 10x lower cost. Fully compatible with Mem0 API. Migrate in 5 minutes without any code changes. Self-host for free. Cloud version available at deepmem.dev.

Overview: Few-Shot Diffusion Models (FSDM) is a framework for few-shot generation using conditional denoising diffusion probabilistic models (DDPMs). This paper presents FSDMs, which adapt the generative process conditioned on a small set of images from a given class to generate samples from previously unseen classes. Key features: - Few-shot generation - Conditional DDPMs - ViT encoder for image patch aggregation - Aggregation methods: mean, lag, cls, sum_patch_mean - Training on small sets of dimension 2-20 - Disjoint classes by default for train/val/test sets Tech stack: Python Why it matters: FSDMs offer a novel approach to few-shot generation by leveraging conditional DDPMs and image patch aggregation, making them a promising tool for applications requiring efficient and effective few-shot generation. GitHub stars: 115

Overview: This project provides a real-time Voice Activity Detection (VAD) solution using Python, implementing VAD algorithms for speech processing and transcription. Key features: - Real-time Voice Activity Detection (VAD) - Simple voice bot and live transcription (realtime transcription) services Tech stack: Python Why it matters: This project offers a practical application of VAD for speech processing and transcription, aiding in various speech-based applications and saving on computation and network bandwidth. Repository: hanifabd/voice-activity-detection-vad-realtime Primary language: Python GitHub stars: 114

Overview: This project is a Jax implementation of Language Quantized AutoEncoders, focusing on training and evaluation code for multi-GPU and Google Cloud TPU environments. Key features: - Multi-GPU and TPU support - Training and evaluation code - Experiment launch via command line Tech stack: Python Why it matters: This implementation provides a practical tool for researchers and developers interested in language quantization and autoencoders, offering efficient training on high-performance hardware. Repository: haoliuhl/language-quantized-autoencoders GitHub stars: 114

Overview: DialOp is a decision-oriented dialogue environment designed for collaborative language agents, focusing on human-AI collaboration. It includes three game modes: Optimization, Planning, and Mediation, each with its own text-based interface and web app. Key features: - Text-based game interfaces for decision-making - Web app interfaces for human-human collaboration - Games associated with the paper "Decision-Oriented Dialogue For Human-AI Collaboration" Tech stack: Python Why it matters: DialOp provides a practical framework for human-AI collaboration, enhancing the interaction between humans and language models through decision-oriented dialogue environments. Repository: jlin816/dialop Primary language: Python GitHub stars: 114 GitHub description: DialOp: Decision-oriented dialogue environments for collaborative language agents README excerpt: # 📠 The DialOp Environments _Collaborative decision-oriented dialogue environments for humans and LLM agents._ These environments are associated with the paper **Decision-Oriented Dialogue For Human-AI Collaboration**. [[Paper]](https://arxiv.org/abs/2305.20076) [[Website]](https://collaborative-dialogue.github.io/)  # Play against GPT-3 Get started and play the games with GPT-3 as an assistant: ``` pip install -e . vim dialop/.api_key python dialop/play_gpt3.py --game {optimization, planning, mediation} ``` # Overview <p align="center"> <img

Overview: This project introduces PairCoder, a novel framework for code generation using LLMs, specifically designed for pair programming. It leverages the collaboration of two agents to propose, select, and refine solutions in an iterative process. Key features: - Integrates a Navigator and a Driver for high-level planning and specific implementation. - Implements multi-plan exploration and feedback-driven refinement to simulate pair programming. - Developed on Ubuntu 16.04 LTS and requires Python 3.10 for environment setup. Tech stack: Python Why it matters: PairCoder offers a novel approach to code generation that mimics the collaborative nature of pair programming, potentially improving efficiency and quality in software development.

Overview: This repository contains the complete codebase for training and evaluating energy-based diffusion models for molecular dynamics simulations. Our approach enables a single model to perform both independent sampling via diffusion denoising and continuous molecular dynamics simulations through a Fokker-Planck-based regularization scheme. Key features: - Energy-based diffusion models for molecular dynamics simulations - Stable, self-consistent scores near the data distribution - Independent sampling via diffusion denoising - Continuous molecular dynamics simulation Tech stack: Python Why it matters: This framework allows for a unified model to perform both generative sampling and accurate energy estimation, making it suitable for applications in materials science, drug discovery, and more. Repository: noegroup/ScoreMD GitHub stars: 113

### Overview: Satori: Reinforcement Learning with Chain-of-Action-Thought Enhances LLM Reasoning via Autoregressive Search ### Key features: - Autoregressive Search Capabilities - Chain-of-Action-Thought (COAT) Reasoning - Transferability to Unseen Domains ### Tech stack: Python ### Why it matters: Satori-Qwen-7B achieves state-of-the-art reasoning performance through a combination of Format Tuning (FT) and reinforcement learning (RL), enabling self-reflection and self-exploitation of alternative strategies without external guidance.

### Overview: This repository provides the official PyTorch implementation of the paper "Deciphering Cross-Modal Alignment in Large Vision-Language Models with Modality Integration Rate" by Qidong Huang et al., accepted at ICCV 2025. The paper presents a novel method for understanding cross-modal alignment in large vision-language models, which is crucial for improving model performance in various applications. ### Key features: - **Modality Integration Rate (MIR)**: A metric to evaluate cross-modal alignment. - **Cross-Modal Alignment**: A technique to align visual and textual information in large models. - **Modality Mask**: A mechanism to compute the alignment between different modalities. ### Tech stack: Python ### Why it matters: This project is significant because it provides a practical solution for understanding and improving cross-modal alignment in large vision-language models, which is essential for enhancing the performance of these models across various applications, including image captioning, visual question answering, and more.

Overview: This repository contains a PyTorch implementation of the Transformer Transducer model, which is a streamable speech recognition model that uses Transformer encoders and RNN-T loss. Key features: - Streamable speech recognition model - Separation of audio and label encoders - Use of Transformer encoders instead of LSTM encoders in RNN-T architecture Tech stack: Python Why it matters: This model offers a more efficient and accurate speech recognition solution compared to traditional LSTM-based models, making it particularly useful for real-time applications.

Overview: Voyage AI's Python library provides cutting-edge embedding and rerankers for semantic search and retrieval-augmented generation (RAG) applications. Key features: - Embedding models convert unstructured data into dense numerical vectors. - Rerankers output relevance scores for queries and documents. Tech stack: Python Why it matters: The library's state-of-the-art embedding models and rerankers enhance semantic search and retrieval accuracy, making it a valuable tool for developing domain-specific or company-specific chatbots and AI applications.

Overview: This project is a TensorFlow 2.0 implementation of the Speech Transformer model, designed for automatic speech recognition. It uses a single neural network to convert acoustic features directly into character sequences. Key features: - Transformer architecture - 2DAttention for feature extraction - Librispeech dataset for training Tech stack: Python Why it matters: This model offers a more efficient and accurate approach to speech recognition compared to traditional methods, potentially improving the quality and speed of speech-to-text conversion. Repository: xingchensong/Speech-Transformer-tf2.0 Primary language: Python GitHub stars: 114 GitHub description: transformer for ASR-systerm (via tensorflow2.0)

### Overview: HiFi-Inpaint is an innovative framework designed to generate high-fidelity human-product images by leveraging reference-based inpainting techniques. This project addresses the challenge of preserving product details in human-product images, which are crucial in various industries such as advertising, e-commerce, and digital marketing. ### Key Features: - **Shared Enhancement Attention (SEA)**: Fine-tunes product features to ensure high-fidelity detail preservation. - **Detail-Aware Loss (DAL)**: Enforces precise pixel-level supervision using high-frequency maps for accurate guidance. - **HP-Image-40K Dataset**: A curated dataset of 40,000 high-quality images for training and testing. ### Tech Stack: Python ### Why It Matters: HiFi-Inpaint provides a robust solution for generating high-fidelity human-product images, addressing the limitations of existing methods in preserving product details, focusing on product detail preservation, and achieving precise guidance through coarse supervision. The project's success is evident from its state-of-the-art performance and the detailed dataset it utilizes, making it a valuable contribution to the field of image inpainting and human-product image generation.

Overview: This repository offers a comprehensive guide to Advanced Retrieval-Augmented Generation (RAG) using the Langchain framework, perfect for enhancing Large Language Models (LLMs) with rich, contextual knowledge. Key features: - Basic RAG - Advanced RAG Techniques - Multi Query Retriever - Self-Reflection-RAG - Agentic RAG - Adaptive Agentic RAG - Corrective Agentic RAG - LLAMA 3 Agentic RAG Local Tech stack: Jupyter Notebook Why it matters: This repository provides practical notebooks for developers to enhance their LLMs with rich, contextual knowledge, making the language models more informed and accurate in their natural language generation.

Overview: Agents-SDK is a portable, high-performance C++ framework designed for developers to create local-first AI agents that can reason, plan, and act independently without relying on the cloud. Key features: - Modular Architecture - Multi-Provider Support (OpenAI, Anthropic, Google, Ollama/llama-cpp) - Optimized for Speed and Memory - Built-In Workflow Patterns (Prompt Chaining, Routing, Parallelization, Orchestrator-Workers, Evaluator-Optimizer) - Autonomous Agents (ReAct, CoT, Plan and Execute, Zero-Shot, Reflexion) - Extensible Tooling System Tech stack: C++ Why it matters: This framework simplifies the development of AI agents, enabling developers to create intelligent systems that can reason, plan, and act locally, enhancing the performance and efficiency of AI applications in various environments.

### Overview: Welcome to the World of AI repository, a platform dedicated to AI-based projects for developers at all skill levels. This open-source repository offers a comprehensive collection of projects from beginners to experts, designed to provide hands-on experience in real-world AI applications. Join the community and contribute to the development of AI-based solutions. ### Key Features: - **Streamlit Projects**: AI projects that can be easily deployed using Streamlit, offering a user-friendly interface. - **Intermediate DL Projects**: Intermediate-level projects focusing on image and audio processing, suitable for developers looking to deepen their knowledge in deep learning. ### Tech Stack: Jupyter Notebook ### Why it Matters: This repository serves as a valuable resource for developers interested in AI and deep learning. It provides a structured way to learn and contribute to real-world AI projects, making it easier to understand and apply AI concepts in practical scenarios. ### Repository: adithya-s-k/World-of-AI GitHub Stars: 113 ### Primary Language: Jupyter Notebook ### GitHub Description: WORLD OF AI: An open-source repository for AI-based projects 🚀, from beginner to expert level, helping contributors start their journey in Artificial Intelligence and Deep Learning. Our projects provide hands-on experience to real-world problems👨‍💻. Join our community and contribute to the development of AI-based solutions 👥. ### README Excerpt: <div align="center"> <h1>WORLD OF AI 🌏</h1> <h3>Website for World of AI Repo

Overview: This repository offers a collection of resources for working with generative AI models on Amazon SageMaker and Amazon Bedrock, including fine-tuning foundation models, building RAG applications, creating agents, and implementing responsible AI practices. Key features: - Building RAG workflows with SageMaker and Bedrock - Advanced model customization on SageMaker AI - DIY agents with SageMaker and Bedrock - Fine-tuning with SageMaker AI and Bedrock - Partner AI apps with SageMaker AI - Amazon Nova customization on SageMaker AI - Serverless model customization with SageMaker AI Tech stack: AWS services including Amazon SageMaker, Amazon Bedrock, Amazon Nova, and Amazon SageMaker AI. Why it matters: This repository provides practical examples and workshops for developers looking to work with generative AI models on Amazon's cloud services, offering a comprehensive approach to fine-tuning models, building applications, and implementing responsible AI practices.

### Overview: This repository contains 82 ready-to-deploy agents for Microsoft Copilot Chat, designed for users on the Copilot Chat tier of the M365 suite. These agents are created using Copilot Studio and can be easily integrated into any commercial M365 environment without requiring a premium Copilot license. The agents are instruction-only, no SharePoint knowledge sources required, and work from pasted input or web search results. They are designed to be used with any commercial M365 license, including Copilot Chat, and do not require coding skills. ### Key Features: - **Instruction-Only**: No SharePoint knowledge sources required. - **No M365 Data Grounding**: Everything works from pasted input or web search results. - **No Premium Copilot License**: Deploys on any commercial M365 license, including Copilot Chat. - **No Coding**: Paste the instruction block into the Copilot Studio builder and you are done. ### Tech Stack: The agents are built using the Copilot Studio, a tool provided by Microsoft for creating custom agents for Copilot Chat. ### Why It Matters: These agents are designed for users on the Copilot Chat tier of the M365 suite, providing a capable AI assistant with a custom agent builder, web search, and file analysis. They are designed to be used with any commercial M365 license, including Copilot Chat, and do not require coding skills. The agents are ready-to-deploy, making them an excellent solution for organizations evaluating which users need premium Copilot licenses.

### Overview: A curated list of AI-built games, featuring playable titles built with AI coding agents like Claude Code, Cursor, Codex, Gemini, Grok, and Lovable. These games leverage AI to create engaging experiences, showcasing the versatility of AI in game development. ### Key Features: - **Games**: A diverse selection of games ranging from action and shooters to puzzle and retro titles. - **Tools & Frameworks**: Built with a variety of tools and frameworks, including Three.js, Phaser, Godot, and more. - **Directories**: Contains directories for different genres and tools. ### Tech Stack: Three.js ### Why it Matters: These games highlight the potential of AI in creating immersive and engaging experiences, demonstrating how AI can be integrated into game development to enhance gameplay mechanics and storytelling.

Overview: AI Study Assistant is an app designed to enhance your learning experience and boost academic performance. It includes features such as an intelligent AI chatbot, lecture summarizer, essay writer, and question generator powered by Google Gemini Pro. Key features: - Personal Tutor: Interact with an AI chatbot for general questions or specific subject explanations. - Lecture Summarizer: Summarize lecture notes or PDF files into a concise PDF format. - Essay Writer: Generate a well-structured essay on a given topic. - Questions Generator: Generate multiple-choice questions with detailed explanations for testing knowledge. Tech stack: Kotlin, Jetpack Compose, Clean Architecture, MVVM Pattern, Ktor, Room DB, Hilt, Preferences DataStore, Kotlin coroutines, Kotlin Flows, Jetpack Compose, Material 3. Why it matters: AI Study Assistant provides a comprehensive set of tools to help you learn and study more effectively, making it easier to understand complex topics, summarize lectures, write essays, and test your knowledge with practice questions. GitHub stars: 113 Repository: mhss1/AIStudyAssistant