Coding & Development

Abnormal - Cloud Email Security is an AI-native security platform designed to protect organizations from advanced email threats such as credential phishing, business email compromise, and account takeovers. Leveraging a unique Behavioral AI, the platform develops a superhuman understanding of normal human behavior within an organization to detect and neutralize anomalies in milliseconds, without human intervention. It offers comprehensive, multi-layered security for cloud email platforms, including inbound email security, email account takeover protection, and security posture management for Microsoft 365. Additionally, Abnormal provides AI Security Agents to automate repetitive SOC workflows and offers SaaS security for applications like Slack and Zoom. The platform integrates easily via a one-click API, ingesting thousands of behavior signals to power its detection engine.

AiderDesk is an open-source, AI-powered development platform designed for professional software engineers, offering transparent and steerable AI orchestration. It goes beyond a simple GUI for Aider, providing a comprehensive orchestration layer that allows users to control AI behavior, review every change, isolate experiments, and extend functionalities. Key features include managing multiple codebases simultaneously with project and task management, utilizing isolated Git worktrees for parallel task development, and advanced task and history control to refine AI context. The platform also boasts a smart context and memory engine, rich review and approval gates for proposed changes, and multi-model orchestration supporting various LLM providers. AiderDesk emphasizes flexibility, integrating seamlessly with existing IDEs and Git workflows, and is highly extensible through custom logic injection and a Model Context Protocol (MCP).

Keel Studio provides a comprehensive solution for rapidly building and deploying production-ready backends. Users can describe their system with a prompt and receive a complete backend featuring a database, APIs, authentication, permissions, observability, and internal tools. The platform is built on a code-first approach, utilizing the Keel Schema as the source of truth for data models, API endpoints, authentication, and secrets. It supports collaboration by instantly creating tools that allow entire teams, including customer service, marketing, and logistics, to interact with data securely. Keel Studio is designed for fast-moving operations teams, enabling them to build custom operational systems like CRM, inventory, and procurement without the complexity of traditional ERPs.

ArcadeDB is a high-performance Multi-Model Database Management System (DBMS) created by Luca Garulli, the founder of OrientDB. Built from scratch with a new engine, ArcadeDB is designed to process millions of records per second with minimal resource usage. It reuses and heavily modifies OrientDB's SQL engine, leveraging low-level Java APIs for advanced mechanical sympathy and reduced Garbage Collector pressure. ArcadeDB is a fully transactional DBMS supporting ACID transactions, structured and unstructured data, a native graph engine, full-text indexing, geospatial querying, and advanced security. It supports various models including Graph, Document, Key/Value, Search Engine, Time Series, and Vector Embedding, and understands multiple languages like SQL, Neo4j Cypher, Apache Gremlin, and MongoDB Query Language. It can be used embedded, remotely via HTTP/JSON, or with Postgres, Redis, and MongoDB drivers.

deep-ctr-prediction is an open-source GitHub repository dedicated to deep learning models for click-through rate (CTR) prediction. It provides implementations of several popular deep neural network (DNN) models, including Wide & Deep, DeepFM, ESMM, Deep Interest Network, ResNet, xDeepFM, AFM, Transformer, and FiBiNET. The code is built using the TensorFlow Estimator API, ensuring compatibility with industrial-grade applications. It utilizes tfrecord format for data storage and the tf.Dataset API for accelerated I/O. A key feature is the separation of feature engineering and model definition, allowing users to easily modify input functions for feature engineering and model functions for model architecture. The repository also supports data storage in Hadoop, with options for local storage.

TensorFlow Decision Forests (TF-DF) is a powerful open-source library designed to integrate state-of-the-art Decision Forest models directly into the TensorFlow ecosystem. It enables users to train, serve, and interpret models such as Random Forests and Gradient Boosted Trees for tasks like classification, regression, and ranking. TF-DF is built upon Yggdrasil Decision Forest (YDF), a high-performance C++ library, ensuring compatibility between models trained in TF-DF and YDF. While TF-DF is available on Linux and Mac, Windows users can access its functionalities via WSL+Linux. The project encourages migration to YDF for enhanced functionality and speed, providing a robust solution for machine learning practitioners.

Helion is a Python-embedded domain-specific language (DSL) designed for authoring machine learning kernels, compiling down to Triton for performant GPU programming. It aims to raise the abstraction level compared to Triton, making it easier to write correct and efficient kernels while enabling more automation in the autotuning process. Helion significantly reduces manual coding effort by evaluating hundreds of potential Triton implementations generated from a single Helion kernel, leading to better performance portability across different hardware. Key features include automated tensor indexing, masking, grid size determination, implicit search space definition, kernel argument management, looping reductions, and various automated optimizations like PID swizzling and loop reordering. It integrates seamlessly with PyTorch operators, allowing users familiar with PyTorch to quickly adopt Helion.

PhyCV is the first Physics-inspired Computer Vision Python library developed by Jalali-Lab at UCLA. It introduces a new class of computer vision algorithms that simulate the propagation of light through physical mediums with diffractive properties, followed by coherent detection. Unlike traditional empirical algorithms, PhyCV leverages physical laws as blueprints, making these algorithms potentially implementable in real physical devices for fast and efficient computation. The library currently includes Phase-Stretch Transform (PST) for edge and texture detection, Phase-Stretch Adaptive Gradient-field Extractor (PAGE) for directional edge detection, and Vision Enhancement via Virtual diffraction and coherent Detection (VEViD) for low-light and color enhancement. Both CPU and GPU versions are available for each algorithm, with GPU versions depending on PyTorch and torchvision.

Gitya is an AI-powered GitHub assistant designed to supercharge development workflows by automating routine tasks and streamlining pull request processes. It seamlessly integrates with existing GitHub workflows as a GitHub App, leveraging advanced AI to manage and automate minor tasks. Key features include AI-enhanced PR management with automated reviews and responses, and ticket automation where labeling a ticket with 'gitya' triggers AI handling of minor tasks. This tool aims to reduce time spent on bug fixes and minor requests, allowing developers to concentrate on high-impact engineering and accelerate project timelines.

Try Gorilla is an AI code assistant hosted on Hugging Face Spaces, designed to help users automate various coding tasks and generate code snippets. This tool is particularly useful for software developers and AI engineers who are looking to streamline their code creation process. While the current live website indicates a runtime error, suggesting it may not be fully operational at this moment, its intended purpose is to provide assistance in coding. The tool aims to simplify the development workflow by offering AI-powered support for generating and managing code.

ONNC (Open Neural Network Compiler) is a retargetable compilation framework specifically engineered for proprietary deep learning accelerators. Its architecture facilitates easy porting to any Deep Learning Accelerator (DLA) design that supports ONNX (Open Neural Network Exchange) operators. ONNC ensures executability across diverse DLAs by converting ONNX models into DLA-specific binary forms, utilizing ONNX's intermediate representation (IR) design and efficient algorithms to minimize data movement overhead. Notably, ONNC is the first open-source compiler available for NVDLA-based hardware designs, capable of compiling models into executable NVDLA Loadable files. Integrating ONNC with the NVDLA software stack empowers developers and researchers to explore NVDLA-based inference design at a system level.

RamaLama is an open-source developer tool designed to simplify the local serving and use of AI models for inference. It leverages familiar OCI containers, allowing engineers to apply container-centric development patterns to AI use cases. The tool eliminates the need for complex host system configurations by automatically detecting GPUs and pulling appropriate accelerated container images. RamaLama supports multiple AI model registries, including OCI Container Registries, HuggingFace, and Ollama, treating models similarly to how Podman and Docker handle container images. It enables secure model execution in rootless containers with no network access by default, ensuring data privacy and temporary data removal upon exit. Users can interact with models via REST API or as a chatbot.

ReconAIzer is a powerful Jython extension designed for Burp Suite, integrating OpenAI (GPT) to significantly optimize the reconnaissance process for bug bounty hunters. This extension automates various tasks, making it faster and easier for security researchers to identify and exploit vulnerabilities. Key functionalities include discovering endpoints, parameters, URLs, and subdomains. Once installed, ReconAIzer adds a contextual menu and a dedicated tab within Burp Suite to display results, streamlining the analysis workflow. Users need to configure their OpenAI API key to utilize its full potential, making it a valuable asset for those looking to leverage AI in their security research.

PipeCNN is an OpenCL-based FPGA Accelerator specifically designed for large-scale Convolutional Neural Networks (CNNs). It leverages High Level Synthesis (HLS) tools to facilitate the design and implementation of customized circuits on FPGAs, significantly speeding up the hardware development cycle compared to traditional RTL-based methodologies. The project provides a generic, yet efficient, OpenCL-based CNN accelerator that is scalable in both performance and hardware resources, making it suitable for various FPGA platforms. PipeCNN supports both Intel OpenCL SDK and Xilinx Vitis based FPGA design flows and includes a ModelZoo with pre-quantized models for networks like VGG-16 and ResNet-50. While the performance may not match the latest state-of-the-art designs, PipeCNN serves as a complete and valuable resource for learning about Deep Learning Architecture (DLA) and experimenting with new ideas in FPGA acceleration.

Jina-Serve is a robust, open-source framework designed for building and deploying multimodal AI applications using a cloud-native stack. It facilitates communication via gRPC, HTTP, and WebSockets, allowing developers to scale their AI services efficiently from local development environments to full production. Key features include native support for major ML frameworks and data types, high-performance service design with scaling, streaming, and dynamic batching, and LLM serving with streaming output. Jina-Serve also offers built-in Docker integration, an Executor Hub, and one-click deployment to Jina AI Cloud, making it enterprise-ready with Kubernetes and Docker Compose support. It provides advantages over tools like FastAPI through DocArray-based data handling, native gRPC support, and seamless microservice scaling.

Tengine, developed by OPEN AI LAB, is a high-performance, modular inference engine specifically designed for embedded devices. It facilitates the rapid and efficient deployment of deep learning neural network models across various AIoT applications. The core modules are developed in C language, with deep framework trimming to suit the limited resources of embedded systems. Tengine features a completely separated front-end and back-end design, which simplifies the porting and deployment to heterogeneous computing units like CPUs, GPUs, and NPUs, thereby reducing evaluation and migration costs. It supports various models and offers tools for conversion and quantization, making it a versatile solution for AI deployment on edge devices.

Dilation is an open-source project that implements dilated convolution for semantic image segmentation. It focuses on multi-scale context aggregation, a technique detailed in its ICLR 2016 conference paper. The repository includes network definitions and pre-trained models, allowing users to segment images using vanilla Caffe. For those interested in training their own models, comprehensive documentation is provided. The project also highlights that dilated convolution is implemented in other deep learning packages like Torch and Lasagne, offering flexibility for developers. It serves as a foundational resource for researchers and developers working on advanced image segmentation tasks.

shell_gpt is a powerful command-line productivity tool that leverages AI large language models like GPT-4 to help users accomplish tasks faster and more efficiently. It eliminates the need for external resources by generating shell commands, code snippets, and documentation directly from the terminal. The tool supports various operating systems including Linux, macOS, and Windows, and is compatible with major shells such as PowerShell, CMD, Bash, and Zsh. Users can interact with shell_gpt through direct prompts, stdin, or even integrate it into their shell for quick completions. It also features a chat mode for conversational interactions and a REPL mode for interactive sessions, allowing for iterative improvements and context preservation.

sisi is a free, open-source command-line interface (CLI) tool designed for semantic image search. It enables users to perform image searches locally on their machines, eliminating the need for external APIs. The tool is powered by node-mlx, a machine learning framework built for Node.js, and leverages the CLIP model to compute image embeddings. sisi supports Macs with Apple Silicon and x64/arm64 Linux, though Windows support is not yet available. It allows users to build and update image indexes for specified directories, list indexed directories, remove indexes, and search for images using natural language queries or image URLs/local files. The indexing process can be time-consuming for large collections without GPU support, but subsequent updates are faster as it only processes new or modified files.

VulcanSQL is an open-source Analytical Data API Framework designed to simplify the creation of RESTful APIs from various data sources like databases, data warehouses, and data lakes. It addresses common pain points in traditional API development, such as time-consuming custom coding, integration complexity, security concerns, and scalability issues. By allowing users to insert variables into templated SQL, VulcanSQL generates SQL statements on the fly, making data accessible for AI agents and data applications. It utilizes DuckDB as a caching layer to boost query speed and reduce API response times. The framework supports flexible deployment options, including Docker, and offers features like OpenAPI document generation for standardization, ensuring easier integration and maintenance.

TokenFormer is the official implementation of the ICLR2025 Spotlight paper, "TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters." This tool introduces a fully attention-based neural network that unifies token-token and token-parameter interactions, maximizing the flexibility of neural network architectures. By tokenizing both data and model parameters, TokenFormer inherently enhances model scalability, allowing for progressively efficient scaling. The architecture is designed to be natively scalable, leveraging attention mechanisms for interactions between input tokens, and between tokens and model parameters. This approach aims to offer greater flexibility than traditional Transformers, contributing to advancements in foundation models, sparse inference (MoE), parameter-efficient tuning, device-cloud collaboration, and vision-language applications.

LatticeWork is a cloud and AI innovations company dedicated to making cutting-edge technology accessible to everyone. Through its Amber brand, LatticeWork provides consumer-focused solutions that offer the convenience of cloud services while prioritizing privacy and freedom. Amber products, such as Amber X and AmberPRO, enable individuals, families, and small businesses to host their own private cloud for media, photo storage, and data management, freeing up space on mobile devices. For businesses, the VAISense line offers hardware, software, and cloud infrastructure to deploy AI at the edge, processing data where it's gathered for faster, more reliable results and enhanced privacy protection. VAISense solutions cater to various industries, including public safety, healthcare, construction, and retail, providing powerful insights through visual AI processing and security tools like OptiView, Security, and Track.

Transformer-TTS is a PyTorch implementation of the "Neural Speech Synthesis with Transformer Network," designed for efficient and high-quality speech synthesis. This model boasts training speeds 3 to 4 times faster than well-known seq2seq models such as Tacotron, while maintaining comparable synthesized speech quality. It utilizes a post-network based on the CBHG model from Tacotron and converts spectrograms into raw audio waves using the Griffin-Lim algorithm. The project includes detailed instructions for data preparation, training the autoregressive attention network and post-network, and generating TTS samples, making it a valuable resource for researchers and developers in speech synthesis.

vlmrun-hub is a comprehensive, open-source repository offering pre-defined Pydantic schemas specifically designed for extracting structured data from unstructured visual domains like images, videos, and documents. It is built for Vision Language Models (VLMs) and optimized for real-world use cases, simplifying the integration of visual ETL into various workflows. The hub addresses the common challenge of VLMs lacking strongly-typed, validated outputs for automation by providing schemas that ensure data conforms to expected types and structures, eliminating complex parsing and validation. Key benefits include ease of use, automatic data validation, type-safety, model-agnostic compatibility, and optimization for visual ETL across industries such as healthcare, finance, and retail.