Projects

Overview: Advising App® by Canyon GBS® is an AI-powered conversational assistant designed specifically for colleges and universities to enhance student engagement and drive institutional success. Key features: - Student Messaging - AI Chat - Guided Outreach Tech stack: PHP Why it matters: Advising App® simplifies communication, provides AI support, and integrates with campus systems for seamless student engagement and relationship management. Repository: canyongbs/advisingapp

Overview: MemN2N is an implementation of End-To-End Memory Networks using Tensorflow, designed for tasks from the bAbl dataset. Key features: - Uses a sklearn-like interface - Supports multiple tasks - Implements stochastic gradient descent with an annealed learning rate schedule Tech stack: Python with Tensorflow Why it matters: MemN2N is a significant contribution to the field of neural network research, offering a practical solution for end-to-end memory network tasks. Repository: domluna/memn2n Primary language: Python GitHub stars: 340 GitHub description: End-To-End Memory Network using Tensorflow README excerpt: Implementation of [End-To-End Memory Networks](http://arxiv.org/abs/1503.08895) with sklearn-like interface using Tensorflow. Tasks are from the [bAbl](http://arxiv.org/abs/1502.05698) dataset.

Overview: PALM-E is an open-source implementation of the SOTA multi-modality foundation model "PALM-E: An Embodied Multimodal Language Model". It is a single large embodied multimodal model that can address a variety of embodied reasoning tasks using multiple observation modalities and multiple embodiments, showcasing positive transfer across diverse joint training domains. Key features: - Single large embodied multimodal model - Addresses a variety of embodied reasoning tasks - Uses multiple observation modalities - Supports multiple embodiments - Exhibits positive transfer across diverse domains Tech stack: Python Why it matters: PALM-E represents a significant advancement in the field of multi-modal AI, offering a comprehensive solution for embodied reasoning tasks that can benefit various industries and applications. Its ability to leverage multiple observation modalities and multiple embodiments makes it highly versatile and adaptable, potentially revolutionizing the way we approach automation and intelligent systems.

Overview: This project provides Rust code that translates the text from Sebastian Raschka's "Build an LLM from Scratch" book, offering a comprehensive guide to building a GPT-style language model. Key features: - Rust translation of PyTorch code - 7 chapters covering understanding large language models, working with text data, coding attention mechanisms, implementing a GPT model, pretraining with unlabeled data, fine-tuning for classification, and fine-tuning for following instructions - Compatible with datasets used in the book - Supports both PyTorch and Rust environments - Documentation available in the project's docs Tech stack: Rust, Candle (a minimalist ML Framework) Why it matters: This project serves as a practical and educational resource for developers looking to build language models using Rust, offering a clear and step-by-step guide based on the detailed instructions provided in Sebastian Raschka's book.

Overview: BehaviorTree.js is an open-source JavaScript implementation of Behavior Trees, which are useful for creating and managing AI behaviors in games and applications. Key features: - Sequences - Selectors - Tasks Tech stack: JavaScript Why it matters: Behavior Trees provide a structured way to implement AI behaviors, making it easier to create complex decision-making systems in various applications. Repository: Calamari/BehaviorTree.js GitHub stars: 339 GitHub description: An JavaScript implementation of Behavior Trees.

### Overview: OpenGraph is a foundation graph model that distills zero-shot graph generalizability from large language models (LLMs). It addresses key technical challenges such as adapting to unseen graph data and leveraging the strengths of LLMs to enhance graph learning performance. ### Key Features: - A unified graph tokenizer for generalizability across different graph properties. - A scalable graph transformer encoder that captures node-wise dependencies. - A data augmentation mechanism enhanced by LLMs to overcome data scarcity issues. ### Tech Stack: Python ### Why It Matters: OpenGraph enables remarkable zero-shot graph learning performance across various settings and domains by effectively distilling zero-shot graph generalizability from LLMs. This approach addresses the limitations of traditional graph learning methods and opens up new possibilities for graph-based applications.

### Overview: This repository implements the paper "ECG arrhythmia classification using a 2-D convolutional neural network," which classifies ECG signals into seven categories, including normal and arrhythmia types, using a deep two-dimensional CNN with grayscale ECG images. ### Key features: - **2-D Convolutional Neural Network (CNN) Implementation**: The repository uses a 2-D CNN architecture for ECG arrhythmia classification. - **Grayscale ECG Images**: The ECG signals are transformed into grayscale images for processing. - **Noise Filtering and Feature Extraction**: Traditional noise filtering and feature extraction steps are omitted, as ECG beats are ignored in noise filtering and feature extraction. - **Data Augmentation**: By augmenting ECG images, the model can train with different viewpoints, improving classification accuracy. - **Robustness**: The use of ECG images enhances robustness in the classification process. ### Tech stack: Python ### Why it matters: This implementation is significant as it addresses the limitations of traditional ECG signal processing methods, such as noise filtering and feature extraction, by using ECG images. This approach results in higher classification accuracy and robustness.

### Overview: `Fuzz4All` is a universal fuzzer designed to target a wide range of input languages and features. Leveraging large language models (LLMs), it generates diverse and realistic inputs for any practical language, making it a powerful tool for software security testing. ### Key features: - Universal language support: Fuzz4All can target any language, including those with complex features. - Diverse input generation: It produces a wide variety of inputs to simulate potential vulnerabilities. - Realistic testing: The generated inputs are designed to mimic real-world scenarios, ensuring more effective testing. - Easy configuration: Config files allow for easy customization of parameters, making it simple to adapt to different environments. ### Tech stack: Python ### Why it matters: `Fuzz4All` is significant because it provides a comprehensive solution for software security testing by leveraging the capabilities of large language models. This approach allows for the generation of realistic and diverse inputs, making it highly effective in identifying potential vulnerabilities in various languages and features.

Overview: txtchat is a project that allows developers to build local chat assistants with AI superpowers. Key features: - Integrates with messaging platforms like Rocket.Chat and Mattermost - Provides a set of intelligent agents that can be used with LLMs, small models, or both - Supports Python 3.10+ and txtai Tech stack: Python Why it matters: txtchat empowers developers to automate their workflows with AI, making them more efficient and capable of handling complex tasks.

Overview: This repository contains curated code samples and tutorials for building agents on the Gemini Enterprise Agent Platform, formerly known as Vertex AI. The platform offers a range of foundation models for AI generation, including Gemini, Veo, Lyria, Nano Banana, and Model Garden. Key features: - Code samples and tutorials - Foundation models: Gemini, Veo, Lyria, Nano Banana, and Model Garden - Development framework: Agent Development Kit (ADK), Agents CLI, Agent Studio, and Agent Garden - Protocols & Interoperability: Agent2Agent, Model Context Protocol (MCP), Agent-to-UI (A2UI), Agent Payments Protocol (AP2), and Universal Commerce Protocol (UCP) Tech stack: Python, JavaScript, and various AI models and tools. Why it matters: This repository provides a comprehensive guide for developers to build, deploy, and scale agents on the Gemini Enterprise Agent Platform, offering a seamless workflow and access to cutting-edge AI models.

Overview: Companion Space 是一款本地优先的二次元陪伴学习应用，允许用户创建独立的学习空间，导入资料并与虚拟角色进行文字或实时语音交流。 Key features: - 独立学习空间 - 资料问答与引用 - 文字与实时语音 - 虚拟角色演出 - 可替换的 AI 服务 Tech stack: Python Why it matters: Companion Space 是一个专注于本地优先的二次元陪伴学习应用，旨在提供一个便捷的学习环境，帮助用户在学习过程中与虚拟角色互动，增强学习体验。 Repository: Johnson-Durui/Companion-Space

Overview: Create and customize your own AI influencer using open-source technologies. Key features: - AI Image Generation - Text-to-Speech - Lip Sync Animation - Consistent Characters - Fully Open Source Tech stack: Stable Diffusion, gTTS, SadTalker, OpenAI GPT Why it matters: No paid APIs required for core functionality, making it accessible for free.

Overview: nanoPerplexityAI is a simple, open-source implementation of perplexity.ai, a service that uses Google's search capabilities to generate responses based on user queries without the need for fancy GUIs or LLM agents. It consists of 200 lines of Python code. Key features: - No fancy GUI or LLM agents - 200 lines of Python code - Generates responses using Google search results and LLM API Tech stack: Python Why it matters: This project is simple and intuitive, making it accessible to developers and users alike. It leverages Google's search capabilities to provide relevant and up-to-date information, ensuring that the responses are based on the latest data.

Overview: Yargy is a rule-based facts extraction tool designed for Russian language, utilizing dictionaries and grammatical rules to extract structured information from text. Key features: - Extracts structured information from Russian texts - Supports Python 3.7+, PyPy 3, and depends on Pymorphy2 - Similar to Tomita parser Tech stack: Python Why it matters: Yargy provides a powerful and efficient method for extracting structured data from Russian text, making it valuable for applications that require natural language processing and data extraction. Repository: natasha/yargy Primary language: Python GitHub stars: 339 GitHub description: Rule-based facts extraction for Russian language README excerpt: 

### Overview: STaRK is a benchmark designed to evaluate large language models (LLMs) in the context of structured and textual knowledge bases. It includes a diverse set of queries that require context-specific reasoning, making it suitable for applications such as product search, academic paper search, and biomedicine inquiries. ### Key Features: - **Query Diversity**: STaRK features a wide range of natural-sounding and practical queries that require context-specific reasoning. - **Structured and Textual Knowledge Bases**: It is designed to be applied to knowledge bases that are both structured and textual, covering applications in product search, academic paper search, and biomedicine inquiries. - **Benchmarking Models**: STaRK provides a benchmark for evaluating LLMs in these contexts, ensuring that the models are robust and perform well in real-world scenarios. ### Tech Stack: - **Python**: The primary programming language used for the development and evaluation of the benchmark. ### Why It Matters: - **Real-World Relevance**: STaRK's diverse and context-specific queries make it highly relevant for applications that require understanding and reasoning with structured and textual knowledge. - **Benchmarking Tool**: It serves as a valuable tool for developers and researchers to evaluate the performance of LLMs in practical, real-world scenarios. ### Repository: - **Repository**: `snap-stanford/stark` - **GitHub Stars**: 337 - **GitHub Description**: (NeurIPS D&B 2024) STaRK: Benchmarking LLM Retrieval on Textual and Relational Knowledge Bases

### Overview: The AI-powered legal compliance assistant for alcohol beverage pricing laws uses Alibaba Cloud Bailian AI capabilities to extract, analyze, and explain New York state-level regulations. It integrates RAG + knowledge-graph reasoning to provide a comprehensive solution for regulatory compliance. ### Key Features: 1. **Document Processing Module**: Upload PDF or HTML regulatory documents, extract text, and segment them. 2. **Vectorization & Embedding Module**: Use Alibaba Cloud Bailian Embedding API for text vectorization, build vector indexes, and enhance semantic similarity search. 3. **Knowledge Graph Construction**: Extract legal entities, identify inter-entity relationships, and output a JSON-formatted knowledge graph. 4. **Semantic Retrieval & Q&A (RAG)**: Retrieve regulations based on vector similarity, enhance context with knowledge graphs, and generate professional answers via Qwen-Turbo model. 5. **Compliance Determination Workflow**: Perform multi-factor logical analysis, risk-level evaluation, and identify and recommend violations. 6. **Front-End Interactive Interface**: Upload documents, interact with the system, generate compliance analysis reports, and query knowledge graphs. ### Tech Stack: Python ### Why It Matters: This AI-powered platform automates the complex process of regulatory compliance for alcohol beverage pricing laws, making it easier for businesses to navigate and adhere to New York state regulations.

### Overview: Eidolon-SC 是一款结合《偶像大师 闪耀色彩》原作剧情、角色语音和文字对话的 AI 语音/文字助手。它支持日语原声和中文字幕，提供回忆原作剧情和跨角色共享记忆的功能。用户可以通过 Google 或 GitHub 登录使用，支持 Docker Compose 一键部署。 ### Key features: - **AI 分身**: 以日语原声回答，附带中文字幕。 - **剧情回忆**: 回忆原作中的剧情。 - **跨角色共享记忆**: 记住用户所说过的话，让偶像之间共享回忆。 - **角色服装**: 支持原作角色的换装功能。 - **日程管理**: 提供周视图和日程安排。 - **周期计算**: 根据用户的言行推算经期。 - **互动玩法**: 包括抽奖、送礼、成就和日历功能。 ### Tech stack: - **Frontend**: Next.js - **Backend**: FastAPI - **WebRTC**: Pipecat - **Docker Compose**: 用于一键部署 ### Why it matters: Eidolon-SC 让用户能够以《偶像大师 闪耀色彩》原作中的偶像身份进行实时对话和互动，体验原作剧情和角色之间的互动。它提供了一种全新的偶像互动方式，结合了原作的剧情和角色记忆，为粉丝们提供了一种新的娱乐和交流方式。

Overview: Kwaak is a powerful tool that allows developers to run a team of AI agents locally, improving test coverage, documentation, and code quality. It provides a chat-like terminal interface for interacting with AI agents. Key features: - Runs AI agents in parallel - Improves test coverage and documentation - Supports writing and executing code - Can answer questions about code, find examples, and write code Tech stack: Rust Why it matters: Kwaak enables developers to focus on building cool features while automating repetitive tasks, improving code quality, and enhancing collaboration with AI-driven insights. Repository: bosun-ai/kwaak GitHub stars: 336 GitHub description: Burn through tech debt with AI agents!

Overview: OrbitKV is a high-performance KV cache designed for LLM inference, extending the capabilities of vLLM and SGLang beyond GPU memory. It stores reusable prefixes in DRAM and optional SSD, ensuring they can be restored after GPU eviction or engine restarts. Key features: - Reuse beyond HBM: save prefixes in DRAM and optional SSD, restoring them after GPU eviction or engine restart. - Share cache capacity: connect multiple local engines to one Manager or fetch compatible peer replicas through Mooncake TENT TCP/RDMA. - Rust ownership: native cache and transfer owners manage byte budgets, reservations, and completion; Python connects engine callbacks and layouts. - Local discovery: bounded inventory streams maintain covered indexes; ordinary queries avoid per-key etcd calls and validate source grants. Tech stack: Rust Why it matters: OrbitKV extends the KV cache of vLLM and SGLang beyond GPU memory, serving repeated documents, shared system prompts, and long conversation prefixes. It optimizes performance and resource utilization for LLM inference, making it essential for efficient and scalable inference environments. Repository: feichai0017/orbitkv GitHub stars: 337 GitHub description: High Performance KV Cache for LLM Inference

Overview: Voice AI is a curated learning path designed for developers to build real-time voice AI agents, from setting up speech-to-text and text-to-speech components to integrating with telephony systems. Key features: - Foundations of speech-to-text and text-to-speech pipelines - Frameworks for building agents, including LiveKit Agents and Pipecat - Components for speech recognition, text generation, and turn-taking - Real-time transport layer using WebRTC or telephony - Production-ready evaluation and deployment strategies Tech stack: JavaScript, TypeScript, WebRTC, LiveKit Agents, Pipecat, Node.js, and Telephony systems. Why it matters: This comprehensive guide helps developers efficiently create and deploy voice AI agents, making it easier to integrate voice technology into various applications and industries.