Projects

### Overview: This portfolio showcases advanced AI and data science projects from various honors and internships. It includes capstone projects on CNN pneumonia detection, healthcare EDA, diabetes risk prediction, and AI ethics research. ### Key Features: - **CNN Pneumonia Detection**: A Convolutional Neural Network trained to detect signs of pneumonia in chest X-ray images. - **Healthcare EDA**: An end-to-end EDA project exploring a real-world healthcare dataset. - **Diabetes Risk Prediction**: A machine learning project predicting diabetes risk. - **AI Ethics Research**: An ethics and societal impact of AI capstone project. ### Tech Stack: Python, scikit-learn, TensorFlow/Keras. ### Why it Matters: This portfolio demonstrates hands-on research and applied projects, contributing to academic integrity, data usage, and future publication opportunities.

Overview: Zypher Agent is a minimal yet powerful framework designed for creating AI agents that live within your applications, providing full control over tools, providers, and execution flow. Key features: - **Agent, Not Workflow**: The agent dynamically decides next steps based on LLM reasoning. - **Git-Based Checkpoints**: Track, review, and revert agent changes with built-in checkpoint management. - **Extensible Tool System**: Built-in tools for file operations, search, and terminal commands with support for custom tools. - **Model Context Protocol (MCP)**: Native support for MCP servers with OAuth authentication. - **Multi-Provider Support**: Works with Anthropic Claude and OpenAI GPT models through a unified interface. - **Loop Interceptor System**: Customize agent behavior with extensible post-inference interceptors. - **Production-Ready**: Configurable timeouts, concurrency protection, and comprehensive error handling. Tech stack: TypeScript Why it matters: Zypher Agent simplifies the process of creating AI agents, making it easier to integrate AI into existing applications while ensuring robustness and ease of use.

### Overview: This repository is a comprehensive collection of articles on various programming languages and frameworks, including Rust, Flutter, Golang, Python, Kotlin, and more. It serves as a tech blog, focusing on topics such as AI, blockchain, and system programming. ### Key Features: - 🚀 Rust: System programming - 🔥️ Mojo: AI / System programming - 🎯 Flutter/Dart: App development framework - 🚀 Zig: Embedded programming - 🐍️ Python: Django, Flask - 🐹 Golang: Microservice framework - 💎 Kotlin: Cross-platform application development - 📱 Tauri: Cross-platform development framework - 🎯 Flutter: Mobile development framework - 💎 Kotlin: Cross-platform application development - 💻 Vue.js: Frontend development framework - ✅️ Weex: Frontend UI framework ### Tech Stack: - Rust, Flutter, Golang, Python, Kotlin, Tauri, Weex, Vue.js, Docker, Kubernetes, Blockchain, and more. ### Why It Matters: This repository is a valuable resource for developers looking to stay updated with the latest technologies and frameworks in the tech industry. It covers a wide range of topics, from system programming to blockchain, making it a one-stop destination for tech enthusiasts and professionals alike.

### Overview: Meta Prompting for AI Systems is an official implementation of the paper "Meta Prompting for AI Systems" (https://arxiv.org/abs/2311.11482), which introduces a framework that elevates the reasoning capabilities of large language models (LLMs) by focusing on the formal structure of tasks rather than content-specific examples. This approach is formalized as a functor that maps a category of tasks to a category of structured prompts, ensuring that compositional problem-solving strategies can be systematically decomposed into modular prompt structures. ### Key Features: - **Meta Prompting (MP)**: A framework that provides a high-level, structural template for how to think rather than specific examples of what to think. - **Recursive Meta Prompting (RMP)**: An automated process where an LLM can generate and refine its own prompts, formalized as a monad. - **Token Efficiency**: Improves results on benchmarks like MATH, GSM8K, and Game of 24 with substantial token efficiency gains over traditional few-shot methods. ### Tech Stack: - Python 3.9+ ### Why It Matters: - **Efficiency and Robustness**: By providing a reusable, structured method for tackling an entire category of tasks, Meta Prompting leads to greater efficiency and robustness in problem-solving. - **Token Efficiency**: Achieves improved results on benchmarks with substantial token efficiency gains over traditional few-shot methods. ### Repository: - **Meta Prompting for AI Systems**: Official implementation of the paper "Meta Prompting for AI Systems" (

### Overview: Cybertron is a Go package designed to simplify the integration of NLP technologies into Go applications. It provides an easy-to-use interface for running pre-trained Transformer models, including BERT, ELECTRA, BART, PEGASUS, and MarianMT, on the Hugging Face models repository. Cybertron supports a variety of tasks such as masked language modeling, text classification, token classification, question-answering, text generation, and more. ### Key Features: - **Model Support**: Cybertron supports a wide range of models including BERT, ELECTRA, BART, PEGASUS, and MarianMT. - **Tasks Supported**: It handles tasks like masked language modeling, text classification (sentiment analysis, topic classification, intent detection), token classification (NER, POS tagging), question-answering, text generation (translation, paraphrasing, summarization), relation extraction, and more. - **Server Mode**: Configurable via environment variables or flags, enabling easy deployment and management. - **Library Mode**: Offers several examples to showcase the capabilities of Cybertron, including Machine Translation (English to Italian) and Zero-Shot Text Classification. ### Tech Stack: Go ### Why It Matters: Cybertron simplifies the integration of NLP technologies into Go applications, making it easier for developers to leverage advanced natural language processing capabilities without requiring additional programming languages or complex frameworks. It is particularly useful for applications that need to perform text-related tasks efficiently and accurately. ### GitHub Stars: 331 ### Repository: Repository: nlpodyssey/c

Overview: CBGBench is an official code repository for the protein-molecule complex binding graph benchmark, providing essential tools and resources for researchers and developers. Key features: - Protein-molecule complex binding graph generation - Benchmarking tools for performance evaluation - Visualization tools for complex structures Tech stack: Python Why it matters: CBGBench offers a comprehensive solution for researchers and developers working on protein-molecule complex binding graph benchmarks, ensuring accurate and efficient performance evaluation.

Overview: This repository contains the official implementation of the paper "Improved Techniques for Training Score-Based Generative Models" by Yang Song and Stefano Ermon, focusing on the NCSNv2 method for generative modeling. It introduces new techniques to improve the performance of score-based generative models, allowing them to produce high-resolution images previously unattainable. Key features: - Improved performance of score-based generative models - Scalability to high-resolution images - No need for adversarial training, producing sharp and diverse samples that rival GANs - Extensive documentation and reproducibility Tech stack: Python Why it matters: This implementation provides a significant advancement in the field of generative modeling, offering improved performance and better sample quality compared to previous methods.

Overview: Hyper-RAG is a powerful Retrieval-Augmented Generation method designed to combat LLM hallucinations by leveraging hypergraph-driven retrieval. This project introduces a novel approach that enhances the accuracy of various language models and surpasses existing SOTA methods in specific medical datasets. Key features: 1. **Hypergraph-Driven Retrieval-Augmented Generation (Hyper-RAG)**: This method comprehensively captures both pairwise and beyond-pairwise correlations in domain-specific knowledge to mitigate hallucinations. 2. **NeurologyCorp Dataset Performance**: Demonstrated an average improvement of 12.3% in accuracy over direct LLM use and outperformed Graph RAG and Light RAG by 6.3% and 6.0%, respectively. 3. **Stable Performance with Query Complexity**: Maintained stable performance as query complexity increased, unlike existing methods that declined. 4. **Evaluation Across Diverse Datasets**: Achieved a 35.5% performance improvement over Light RAG using a selection-based assessment. 5. **Lightweight Variant**: Hyper-RAG-Lite, which achieved twice the retrieval speed and a 3.3% performance boost compared to Light RAG. Tech stack: Python Why it matters: Hyper-RAG is a significant advancement in the field of language model reliability and performance, particularly in medical applications where the consequences of hallucinations can be severe. Its ability to enhance the accuracy of LLMs and reduce hallucinations makes it a crucial tool for applications that require high reliability, such as medical diagnostics and decision-making.

Overview: Pyramidal Convolution is a PyTorch implementation of the paper "Pyramidal Convolution: Rethinking Convolutional Neural Networks for Visual Recognition," which presents a novel approach to improving the recognition capabilities of convolutional neural networks (CNNs) on the ImageNet dataset. Key features: - Pyramidal Convolution models achieve higher accuracy compared to baseline models. - The models show significant improvements in recognition accuracy, especially with more complex training settings. Tech stack: The project uses PyTorch, ImageNet dataset, and NVIDIA Docker for a fast alternative without the need for installing deep learning libraries. Why it matters: Pyramidal Convolution offers a promising solution for enhancing the performance of CNNs in visual recognition tasks, potentially leading to better results in various applications where image recognition is crucial.

Overview: AlignProp is a method that aligns large-scale text-to-image diffusion models to downstream reward functions using end-to-end backpropagation of the reward gradient through the denoising process. It is 25x more sample and compute efficient than reinforcement learning methods for finetuning Stable Diffusion. Key features: - AlignProp finetunes low-rank adapter weight modules. - Uses gradient checkpointing to reduce memory usage. - Supports finetuning diffusion models for various downstream tasks such as image-text semantic alignment, aesthetics, compressibility, and controllability of the number of objects present. Tech stack: Python Why it matters: AlignProp achieves higher rewards in fewer training steps than alternatives, making it a straightforward choice for optimizing diffusion models for differentiable reward functions of interest.

### Overview: 本书系统地介绍了多模态大模型的关键技术、基础模型和典型应用，旨在帮助低年级本科生和人工智能领域的从业者更容易理解技术内容。通过深入浅出的方式，本书提供了多模态大模型的结构和技术剖析，以及多个经典应用的实例。本书还探讨了实现通用人工智能的前沿技术，如因果推理、世界模型、具身智能与多智能体等，为读者提供清晰的视角，帮助学术界和工业界更好地理解多模态大模型的技术和新一代人工智能的发展方向。 ### Key features: - **深入浅出的技术介绍**：通过直观实例剖析多模态大模型的关键技术点。 - **经典应用实例**：详细解析视觉问答、AIGC和具身智能等典型应用。 - **前沿技术探讨**：涵盖因果推理、世界模型、具身智能与多智能体等前沿技术。 - **教学与实践结合**：适合高校高年级本科生和研究生使用，教学课时安排灵活，可满足32-56学时的教学需求。 ### Tech stack: HTML ### Why it matters: 本书为学术界和工业界提供了一个清晰的视角，帮助人工智能科研工作者全面了解多模态大模型的技术和新一代人工智能的发展方向。通过系统介绍和实例解析，本书旨在为读者提供清晰的指导，助力人工智能领域的科研和应用发展。

### Overview: Vakyansh Open Source Models is an open-source project dedicated to providing pre-trained and finetuned ASR (Automatic Speech Recognition) models for Indian languages. These models are designed to improve the accuracy and efficiency of speech recognition across various languages, including Hindi, Kannada, Tamil, and more. ### Key Features: - **Pretrained ASR Models:** Models pre-trained using Nemo toolkit and NVIDIA compute resources. - **Vakyansh-Conformer-SSL:** Pre-trained on 34,000 hours of unlabeled audio across 39 Indian languages. - **CLSRIL-23:** Cross-lingual speech representations for 23 Indic languages. - **hindi_pretrained_4kh:** Trained on 4,200 hours of Hindi data. - **kannada_pretrained_1400h:** Trained on 1,400 hours of Kannada data. - **Finetuned ASR Models:** Models fine-tuned on specific language data. - **Hindi:** Pre-trained on Vakyansh Conformer SSL and finetuned on 2,500 hours of Hindi data. - **Indian English:** Pre-trained on Vakyansh Conformer SSL and finetuned on 700 hours of Indian English data. - **Kannada:** Pre-trained on Vakyansh Conformer SSL and finetuned on 1,000 hours of Kannada data. - **

Overview: Fuzzylite is a comprehensive fuzzy logic control library designed for efficient and easy implementation of fuzzy logic controllers in C++. Key features: - Supports Mamdani, Takagi-Sugeno, Larsen, Tsukamoto, Inverse Tsukamoto, and Hybrid controllers. - Offers 5 basic linguistic terms: Triangle, Trapezoid, Rectangle, Discrete, and SemiEllipse. - Provides 8 extended linguistic terms: Bell, Cosine, Gaussian, GaussianProduct, PiShape, SigmoidDifference, SigmoidProduct, and Spike. - Includes 7 edges: Arc, Binary, Concave, Ramp, Sigmoid, SShape, and ZShape. - Features 3 activation methods: General, Proportional, and Threshold. - Supports 9 conjunction and implication (T-Norms) methods: Minimum, AlgebraicProduct, BoundedDifference, DrasticProduct, EinsteinProduct, HamacherProduct, NilpotentMinimum, LambdaNorm. Tech stack: C++ Why it matters: Fuzzylite simplifies the design and implementation of fuzzy logic controllers, making it easier and more efficient for developers to create and operate fuzzy logic systems.

Overview: Pathomic Fusion is an integrated framework that fuses histopathology and genomic data using deep learning techniques, aiming to improve cancer diagnosis and prognosis. Key features: - Utilizes attention gating and tensor fusion for efficient integration of histopathology and -omic data. - Capable of processing both CNNs and GCNs for parameter efficiency. - Adaptable for integrating multiple -omic modalities with histopathology. - Can be used for enhanced diagnostic, prognostic, and therapeutic response determinations. Tech stack: Jupyter Notebook Why it matters: This method simplifies the integration of histopathology and -omic data, making it more accessible and efficient for medical applications. Repository: mahmoodlab/PathomicFusion GitHub stars: 329 GitHub description: Fusing Histology and Genomics via Deep Learning - IEEE TMI README excerpt: # Pathomic Fusion: An Integrated Framework for Fusing Histopathology and Genomic Features for Cancer Diagnosis and Prognosis Overview: We propose a simple and scalable method for integrating histology images and -omic data using attention gating and tensor fusion. Histopathology images can be processed using CNNs or GCNs for parameter efficiency or a combination of the two. The setup is adaptable for integrating multiple -omic modalities with histopathology and can be used for improved diagnostic, prognostic, and therapeutic response determinations. Key features: - Utilizes attention gating and tensor fusion for efficient integration of histopathology and -omic data. - Capable of processing both CNNs and GC

Overview: Masakhane is an open-source research effort focused on building and sharing NLP tools for African languages. Key features: - Data sets and tools for African language NLP research - Community-driven, distributed, and online research platform - Open-source, continuous improvement approach Tech stack: Lua, Python, and other open-source technologies Why it matters: It aims to enhance language preservation, increase global visibility, and contribute to the NLP research landscape by providing practical tools and community-driven insights. Repository: masakhane-io/masakhane-mt GitHub stars: 329 GitHub description: Machine Translation for Africa

Overview: This repository contains a new generative model of chatbot based on seq2seq modeling, designed for fine-tuning on closed domain datasets for real-world applications. Key features: - Fine-tuning of the trained model using a closed domain dataset - Sharing an embedding layer between the encoding and decoding processes through the adoption of a new model - Thought vector encoding the last two utterances of the conversation and concatenation to a dense vector for answer generation - Pseudocode algorithm for training the model Tech stack: Python Why it matters: This model offers improved context sensitivity and avoids the risk of vanishing gradient, making it more effective during training compared to the canonical seq2seq model. Repository: oswaldoludwig/Seq2seq-Chatbot-for-Keras Primary language: Python GitHub stars: 329

### Overview: Go-rag 是一个基于 eino、gf 和 Vue 实现的知识库检索系统，旨在提供高效、便捷的知识库管理功能。通过整合 eino 提供的向量计算能力、gf 实现的知识库构建和 Vue 技术，Go-rag 实现了从创建知识库到文档检索的一站式解决方案。 ### Key features: - **知识库创建与管理**: 能够创建和编辑知识库文档，支持多种文档格式（如 Markdown、PDF 和 HTML）。 - **文档检索**: 实现了文档的自动切割和检索功能，支持长文档的快速检索。 - **多知识库支持**: 可以同时支持多个知识库，方便用户管理和使用。 - **文档编辑**: 支持文档的编辑和修改，方便用户随时更新知识库内容。 - **长文档自动切割**: 自动将长文档分割成更小的 chunk，便于检索和使用。 - **多路召回**: 结合内容向量和问答向量进行召回，提高检索效果。 - **集成 mcp**: 实现了基于 Go-rag 的深度学习模型集成，增强了检索能力。 - **兼容 Windows 和 macOS**: 通过集成到 deepchat，实现了在 Windows 和 macOS 系统上的无缝使用。 ### Tech stack: Go-rag 使用了 Go 语言进行后端开发，结合了 eino、gf 和 Vue 技术实现前端界面。后端主要使用 Docker 和 Docker Compose 进行部署和管理。 ### Why

Overview: HumanMAC is an innovative framework for human motion prediction that addresses the limitations of existing methods by utilizing a denoising diffusion style. It introduces a motion diffusion model that generates motions from random noise, enabling more continuous and controllable predictions. Key features: - Novel denoising diffusion framework for motion prediction - Significantly improved performance compared to traditional encoding-decoding methods - Effective switching of different categories of motions in realistic tasks Tech stack: Python Why it matters: Human motion prediction is a crucial problem in computer vision and graphics with numerous practical applications. HumanMAC's innovative approach offers promising algorithmic properties, making it a significant advancement in the field. Repository: LinghaoChan/HumanMAC Primary language: Python GitHub stars: 328 GitHub description: [ICCV-2023] Official code for work "HumanMAC: Masked Motion Completion for Human Motion Prediction".

### Overview: Tonic Validate is a comprehensive framework designed to evaluate the quality and performance of LLM (Large Language Model) and Retrieval Augmented Generation (RAG) applications. It provides a suite of metrics to assess various aspects of responses, including answer correctness, LLM hallucination, and more. The framework is built on Python and is compatible with PyPI, with 328 GitHub stars, indicating its growing popularity among developers. ### Key Features: - **Tonic Validate Metrics**: A collection of metrics to evaluate the quality of responses, including answer correctness, LLM hallucination, and more. - **Metric Inputs**: Users can input their data to be evaluated, ensuring that the data is of high quality and secure. - **Scoring With Metrics**: The framework scores the responses based on the provided metrics, providing a clear assessment of performance. - **Viewing the Results**: Users can easily view and track the results of their evaluations, making it easier to monitor and improve their RAG applications. ### Tech Stack: - **Python** - The primary programming language used to build and run Tonic Validate. ### Why It Matters: Tonic Validate is essential for developers looking to enhance the performance and reliability of their LLM and RAG applications. By providing a standardized way to evaluate responses, it helps ensure that the outputs are not only accurate but also trustworthy, which is crucial for applications that rely on accurate information retrieval and generation. ### Repository: The repository for Tonic Validate is hosted on GitHub at [TonicAI/tonic_validate](https://github.com/T

### Overview: Apache Texera (Incubating) is an open-source platform designed for human-AI collaborative data science using visual workflows. It allows analysts to construct, execute, and refine data analysis tasks through an intuitive graphical user interface, assisted by AI agents that understand natural-language instructions. Texera is suitable for a variety of applications, including "AI for Science," and supports both local and cloud-based deployments. ### Key Features: 1. **Natural-Language Data Science through AI Agents**: Enables analysts to perform data science tasks using natural language. 2. **Intuitive GUI-Based Workflows**: Facilitates the creation, execution, and refinement of data science workflows. 3. **Real-Time Collaboration**: Supports real-time collaboration for workflow editing and execution. 4. **Runtime Debugging and Interactive Workflow Execution**: Provides tools for debugging and interactive execution of workflows. 5. **Language-Agnostic Workflow Runtime**: Native support for Python and Java, enabling language-agnostic workflows. 6. **Parallel Backend Engine**: Supports scalable big-data processing through a parallel backend engine. 7. **Separation of Compute and Storage**: Allows flexible cloud deployment, separating compute and storage for efficient use. ### Tech Stack: Scala ### Why It Matters: Texera is well suited for a wide range of applications, including "AI for Science," making advanced AI and data science capabilities accessible to a broader community. It can run on a laptop for local use or be deployed in the cloud to support scalable processing of large datasets. ### GitHub Stars: 326 ### Primary Language: Scala ### GitHub Description: