Projects

Overview: This repository serves as a tutorial for the NLP journey, showcasing various techniques and libraries used in natural language processing. Key features: - Data Augmentation - Text Preprocessing (Tokenization, Part of Speech, Lemmatization, Stemming, Stop Words, Phrase Word Recognition, Spell Checking, Machine Translation) - General NLP Tricks Tech stack: Python Why it matters: This repository provides practical examples and resources for understanding and implementing various NLP techniques, making it valuable for developers looking to enhance their NLP skills.

Overview: Qwen2.5-Math series expands on the Qwen2-Math series by introducing CoT and TIR reasoning methods, offering improved performance on Chinese and English math problems. Key features: - Qwen2.5-Math-1.5B/7B/72B: Base models for general math problems. - Qwen2.5-Math-1.5B/7B/72B-Instruct: Instruction-tuned models for better chat performance. - Qwen2.5-Math-RM-72B: Mathematical reward model for additional performance. Tech stack: Python. Why it matters: The Qwen2.5-Math series models have demonstrated significant improvements in performance compared to the Qwen2-Math series on Chinese and English math benchmarks, particularly with the use of CoT and TIR reasoning methods.

Overview: MAXIM is an innovative multi-axis multi-layer perceptron designed for image processing tasks such as denoising, deblurring, deraining, and enhancement. This model achieves state-of-the-art performance across various image processing benchmarks. Key features: - Multi-axis MLP architecture for comprehensive image processing - SOTA performance on deblurring, dehazing, and enhancement tasks - Versatile application across different image processing scenarios Tech stack: Python Why it matters: MAXIM is a groundbreaking model that significantly improves image quality in various applications, making it a valuable tool for researchers and practitioners alike. Its robust performance on benchmark datasets highlights its effectiveness and versatility in image processing tasks.

Overview: zlib is a command-line tool designed for Z-Library, allowing users to search, download, send books to Kindle, and integrate with AI agents. Key features: - Interactive search - Book download - Download history browser - Send to Kindle - Usage profile - Themes - Proxy & custom domain support Tech stack: Go Why it matters: zlib simplifies the process of accessing Z-Library, enabling users to efficiently search, download, and send books to their Kindle devices, making it easier to integrate with AI agents and manage their reading habits.

Overview: This project is a port of the Single Shot MultiBox Detector (SSD) algorithm to the Keras framework, originally developed for the Caffe library. The SSD is a convolutional neural network designed for object detection. Key features: - SSD implementation in Keras - Ported weights from the original models - Available weights for both 300x300 model and an older version with 3x3 convolution in `pool6` Tech stack: Keras, TensorFlow, OpenCV Why it matters: This project provides a practical and efficient way to apply SSD for object detection tasks, making it accessible for developers and researchers alike.

Overview: Datumbox is an open-source Machine Learning framework written in Java, designed to facilitate rapid development of Machine Learning and Statistical applications. It includes a wide range of machine learning algorithms and statistical methods, capable of handling large datasets. Key features: - Support for multiple machine learning algorithms and statistical methods - Ability to handle large datasets - Comprehensive documentation and JUnit tests for public methods and classes - Pre-trained models for sentiment analysis, subjectivity analysis, topic classification, spam detection, etc. - Documentation and examples available for further learning Tech stack: Java Why it matters: This framework provides a powerful and flexible toolset for developers looking to integrate machine learning and statistical capabilities into their projects, making it an essential resource for those working in areas such as natural language processing, data analysis, and predictive modeling.

### Overview: HaGRID is a large image dataset designed for hand gesture recognition (HGR) systems. It contains 1,086,158 FullHD RGB images divided into 33 gesture classes, including a new "no_gesture" class with domain-specific natural hand postures. The dataset is suitable for various applications such as video conferencing, smart home automation, and automotive sectors. ### Key Features: - **HaGRIDv2**: 1.5T dataset with 1,086,158 images divided into 33 gesture classes and a "no_gesture" class. - **Dynamic Gesture Recognition Algorithm**: Introduced to enable dynamic gesture recognition using exclusively static gestures. - **Pretrained Models**: Available in the repository for real-time applications in video conferencing, smart home control, and automotive systems. ### Tech Stack: - Python ### Why It Matters: HaGRID provides a comprehensive dataset for developing robust HGR systems, enhancing their accuracy and applicability across various industries.

Overview: Chatd is a desktop application that allows you to interact with your documents using a local large language model. It ensures that your data stays on your computer and is never sent to the cloud, providing a completely private and secure experience. Key features: - Local large language model (Mistral-7B) for document interaction - Runs on Ollama, a cross-platform LLM server - No need to install additional software - Runs on both macOS, Windows, and Linux - Supports GPU acceleration for enhanced performance - Customizable with a custom model option Tech stack: JavaScript, Ollama, Mistral-7B Why it matters: Chatd offers a secure and private way to interact with your documents, ensuring your data stays local and away from the cloud. The ability to run a local LLM model directly on your computer makes it accessible and efficient, without the need for external servers or cloud storage.

Overview: This repository contains a comprehensive guide to machine learning using TensorFlow 2 and Keras, focusing on practical applications in Python using Jupyter notebooks. The book covers various topics such as building neural networks, handling imbalanced datasets, hyperparameter tuning, and deploying models to production. Key features: - TensorFlow 2 and Keras tutorials - Machine learning fundamentals - Real-world problem-solving examples - Jupyter notebook-based learning Tech stack: TensorFlow 2 and Keras Why it matters: This repository provides a practical, hands-on approach to learning machine learning, making it accessible and useful for developers looking to advance their skills in Python and TensorFlow/Keras. Repository: curiousily/Deep-Learning-For-Hackers GitHub stars: 1079 GitHub description: Machine Learning tutorials with TensorFlow 2 and Keras in Python (Jupyter notebooks included) - (LSTMs, Hyperameter tuning, Data preprocessing, Bias-variance tradeoff, Anomaly Detection, Autoencoders, Time Series Forecasting, Object Detection, Sentiment Analysis, Intent Recognition with BERT)

### Overview: OpenOps is a No-Code FinOps automation platform designed to help organizations reduce cloud costs and streamline financial operations. It provides customizable workflows, a dedicated database, and visualization tools, enabling collaboration between FinOps teams and other stakeholders. ### Key Features: - **Pre-Built FinOps Workflows**: A library of best-practice workflows designed with input from FinOps leaders. - **No-Code Experience**: Approachable for non-technical practitioners, but allows "dropping into" code when needed. - **Flexible Workflow Editor**: Use OpenOps's dedicated No-Code editor to build workflows from scratch or customize existing ones. - **Deep Integrations**: Native connections with cloud providers, databases, analysis tools, communication platforms, and project management systems. - **Human-in-the-Loop Controls**: OpenOps comes with HITL controls for critical approval workflows. - **Workflow Versioning & Traceability**: Test workflow steps, maintain workflow versions, and track every action inside a workflow with logs. - **Centralized Management**: Log and process opportunities with tables that allow approvals, dismissals, false-positive marking, and snoozing. ### Why OpenOps? - **Consolidating Optimization Opportunities**: Surfaces cost-saving opportunities from native and third-party FinOps visibility tools. - **Suggesting Practical Actions**: Provides actionable optimization suggestions. - **Customization**: Enables customization of pre-built optimizations or authoring new ones. - **Integration**: Seamless integration with a broad range of platforms, including cloud providers, databases, FinOps tools, communication platforms, and task management services. ###

### Overview: AIRecon is an autonomous cybersecurity agent that leverages a self-hosted Large Language Model (Ollama) alongside a Kali Linux Docker sandbox and a Textual TUI interface. It automates security assessments, penetration testing, and bug bounty reconnaissance, offering a fully offline and private operation. ### Key Features: - **Backend Agnostic**: Supports any OpenAI-compatible LLM gateway (LiteLLM, vLLM, or local Ollama proxy). - **Reasoning Detection**: Automatically detects and uses the best reasoning model. - **Caido Integration**: Built-in tools for list, replay, automation, findings, scope, sitemap, and intercept. - **Privacy**: Targets, tool outputs, and reports remain local, ensuring privacy. - **Verified Learning**: Compounds only proven findings, avoiding noise. - **Session Resume**: Resumes chat and tool calls for seamless interaction. - **Local Knowledge Base**: Includes 1.09M records, optional for enhanced security knowledge. ### Tech Stack: - Python ### Why It Matters: AIRecon is designed to automate security assessments, penetration testing, and bug bounty reconnaissance without the need for API keys or cloud dependency. It offers a fully offline and private operation, making it ideal for researchers and security professionals looking to enhance their skills and knowledge.

### Overview: This repository is a curated list of resources for large language models used in role-playing with assigned personas. It includes papers and resources related to character role-playing, personalization, multi-agents, and anthropomorphic cognition. The survey paper "From Persona to Personalization: A Survey on Role-Playing Language Agents" was recently accepted to TMLR 2024, highlighting its importance in the field. ### Key features: - **Role-Playing:** Includes papers on characters, demographics, personalization, multi-agents, GUI agents for games, and anthropomorphic cognition. - **Multi Agents:** Focuses on research on large language models and their capabilities. - **GUI Agents for Games:** Discusses how large language models can be used in game environments. - **Anthropomorphic Cognition:** Covers personality traits, social intelligence, and theory-of-mind. ### Tech stack: The repository uses Markdown for documentation and includes links to relevant papers and resources. ### Why it matters: This repository provides a comprehensive overview of the latest research in role-playing language models, making it a valuable resource for developers and researchers interested in this field.

Overview: This repository contains notes and code for the first part of Aurelien Geron's "Hands-On ML with Scikit-Learn, Keras & TensorFlow" book, focusing on machine learning concepts using production-ready Python frameworks like Scikit-Learn, Keras, and TensorFlow. Key features: - Scikit-Learn - Keras - TensorFlow Tech stack: Python Why it matters: The book provides a comprehensive guide to data science, machine learning, and deep learning, focusing on practical examples and hands-on learning. It is suitable for those with basic Python programming knowledge and a foundational understanding of mathematics and statistics.

### Overview: This curated list provides practical resources, libraries, and information sources for text processing in Ruby, focusing on Natural Language Processing (NLP) using the Ruby programming language. It covers various subtasks such as pipeline generation, language identification, segmentation, lexical processing, phrasal level processing, syntactic processing, semantic analysis, and more. ### Key Features: - **Pipeline Generation**: Tools and libraries for creating text processing pipelines. - **Multipurpose Engines**: APIs and engines for various NLP tasks. - **Language Identification**: Detecting the language of a text. - **Segmentation**: Breaking down text into smaller units. - **Lexical Processing**: Stemming, lemmatization, counting types and tokens, and filtering stop words. - **Phrasal Level Processing**: Constituency parsing. - **Syntactic Processing**: Constituency parsing. - **Semantic Analysis**: Semantic analysis. - **Pragmatical Analysis**: Pragmatical analysis. - **Spelling and Error Correction**: Text correction tools. - **Text Alignment**: Aligning text for comparison. - **Machine Translation**: Translation tools. - **Sentiment Analysis**: Analyzing text sentiment. - **Numbers, Dates, and Time Parsing**: Parsing numerical, date, and time data. - **Named Entity Recognition**: Recognizing named entities in text. - **Text-to-Speech-to-Text**: Conversational text-to-speech conversion. - **Dialog Agents, Assistants, and Chatbots**: Tools for building conversational agents. - **Linguistic Resources**: Resources like

### Overview: This repository is a curated collection of papers that have been encountered by the author in their NLP practice and deemed valuable for practitioners. The papers cover various aspects of NLP, including part-of-speech tagging, parsing, named entity recognition, and coreference resolution. ### Key Features: - **Part-of-Speech Tagging**: Demonstrates a powerful HMM-based POS tagger and introduces a number of linguistic features for building a state-of-the-art POS-tagging system. - **Parsing**: Shows that unlexicalized probabilistic context-free grammars can surpass lexicalized PCFGs, and introduces the use of expectation-maximization for learning accurate and compact latent nonterminal symbols for tree annotation. - **Named Entity Recognition**: Demonstrates the effectiveness of seq2seq architectures with attention on constituency parsing, achieving state-of-the-art performance. - **Coreference Resolution**: Proposes a sieve-based approach to coreference resolution that has been SOTA for many years until deep learning approaches became prevalent. ### Tech Stack: The repository uses a curated collection of papers across various NLP tasks. ### Why It Matters: These papers represent significant advancements in NLP and are essential reading for practitioners looking to stay updated with the latest techniques and methodologies in the field.

Overview: UniRepLKNet is a unified ConvNet architecture that combines the strengths of large kernels and universal perception, achieving state-of-the-art performance across multiple modalities, including images, audio, and time-series forecasting. Key features: - A unified architecture that excels in image recognition. - Universal perception ability across multiple modalities. - Achieves ImageNet accuracy of 88.0%, COCO AP of 56.4, and ADE20K mIoU of 55.6 with ImageNet-22K pretraining. - Demonstrates leading performance in audio recognition and global temperature & wind speed forecasting tasks. Tech stack: Python Why it matters: UniRepLKNet leverages the unique characteristics of large kernels to unlock the exceptional performance of ConvNets in diverse domains, showcasing the potential of ConvNets to conquer new territories and adapt to various modalities and tasks.

### Overview: Diffusion Planner is an innovative approach to autonomous driving planning that leverages the power of diffusion models to enhance motion planning without overly relying on refinement. This project was presented at ICLR 2025 as an Oral Presentation, showcasing its significant contributions to the field. ### Key Features: - **DiT-based Architecture**: Focuses on the fusion of noised future vehicle trajectories and conditional information. - **Joint Modeling**: Unifies motion prediction and closed-loop planning as future trajectory generation. - **Fast Inference**: Achieves around 20Hz for real-time performance, making it suitable for high-speed autonomous driving scenarios. ### Tech Stack: - Python ### Why It Matters: Diffusion Planner represents a groundbreaking effort in fully harnessing the power of diffusion models for autonomous driving, offering a more flexible and efficient solution compared to traditional refinement-based methods. This project is significant for its practical applications in real-world autonomous driving tasks, particularly in complex and dynamic environments.

### Overview: AutoResearch is a curated list of use cases and optimization traces for AutoResearch, a prompt designed to optimize code by iteratively improving a single file. The project includes open-source implementations and optimization traces for various tasks, including LLM training, GPU kernel optimization, and voice agent prompt engineering. ### Key Features: - **Use Cases:** A curated list of use cases with optimization traces. - **Open Source Implementations:** Includes open-source implementations for various tasks. - **Community Adaptation:** The loop structure is adaptable to different optimization tasks. ### Tech Stack: AutoResearch is built using Markdown and Python. ### Why It Matters: AutoResearch provides a systematic approach to optimizing code by iteratively improving a single file, making it a valuable tool for developers looking to enhance their code efficiency.

### Overview: pgGraph is an open-source extension for PostgreSQL that allows you to run graph search, traversal, shortest path, and relationship queries directly against your existing PostgreSQL tables. It builds a derived graph index and lets you query it from SQL using functions in the `graph` schema. ### Key Features: - **Graph Database Superpowers**: Enhance your existing Postgres data with powerful graph querying capabilities. - **Stable and Reliable**: pgGraph ensures your data remains the source of truth, providing a stable and reliable solution. - **SQL-Driven Queries**: Use SQL functions to perform graph operations, making it easy to integrate into your existing PostgreSQL ecosystem. ### Tech Stack: Rust ### Why It Matters: pgGraph offers a way to leverage the power of graph databases without the need for a separate database, making it a valuable tool for developers looking to enhance their Postgres data with graph capabilities.

### Overview: Quant-Finance-Resources is a comprehensive collection of courses, articles, and resources designed to help beginners and professionals in the field of quantitative finance. The repository is intended for STEM graduates, focusing on STEM skills and mathematics. The courses cover topics such as numerical linear algebra, probability, reinforcement learning, market microstructures, and advanced financial derivatives. ### Key Features: - **Numerical Linear Algebra for Coders** by fast.ai - **Introduction to Probability** by MIT OCW - **Topics in Mathematics with Applications in Finance** by MIT Mathematics - **Machine Learning** by Tom Mitchell - **Introduction to Reinforcement Learning** by David Silver - **Quantopian Lectures** for Python and Statistics - **Global Financial Crisis** by Timothy F. Geithner - **Introduction to Market Microstructures** by Paul Besson - **Python for Finance** by Yves Hilpisch - **Accelerated C++** by Andrew Koenig and Barbara E. Moo - **Effective Modern C++** by Scott Meyers - **Statistics** by David Freedman, Robert Pisani, Roger Purves - **How to Lie with Statistics** by Darrell Huff - **Thinking Strategically** by Avinash Dixit and Barry Nalebuff - **Analysis of Financial Time Series** by Ruey S. Tsay - **Deep Learning** by Ian Goodfellow et al. - **Reinforcement Learning An Introduction** by Sutton and Barto - **Options, Futures, and other Deriv