Coding & Development

Genoshi is an AI consulting firm specializing in full-stack, battle-tested AI solutions. The company focuses on providing expert guidance and implementation for building explainable and secure AI applications at scale. While specific features are not detailed on the homepage, the emphasis is on consulting services that enable businesses to deploy and manage robust AI solutions with a strong focus on security and transparency. This positions Genoshi as a partner for organizations looking to integrate advanced AI capabilities into their operations, ensuring reliability and understanding of the AI's decision-making processes.

Clarifai is a leading platform for AI compute orchestration, designed to streamline the end-to-end execution of complex AI tasks. It dynamically manages compute resources across cloud, on-premise, and edge environments, enabling users to build, train, and deploy AI models without friction. The platform offers unmatched speed for AI inference and reasoning on GPUs, significantly cutting infrastructure costs by over 90%. Clarifai supports flexible deployments for custom, open-source, and third-party models, including OpenAI-compatible outputs for easy migration. It features AI Runners for connecting local models to the cloud, automated deployments, and Pythonic SDKs, making it ideal for developers and enterprises seeking to optimize AI workloads.

ISS - Intelligent Security Systems Middle East is a global leader in security surveillance technology, specializing in video capture, recording, digital data transmission, and pattern analysis in video and images. The core of their offering is the ISS SecurOS® platform, which functions as a comprehensive security ecosystem. This platform is available as an end-to-end solution and boasts extensive integrations with numerous third-party security systems and devices. ISS provides a wide portfolio of professional, export-oriented video management and video analytics products, backed by significant experience in large-scale projects and collaborations with leading world integrators. They offer full support from design and deployment to post-installation and maintenance, serving industries such as smart/safe cities, facilities, transportation, critical infrastructure, retail, logistics, banking, finance, and law enforcement.

The London Initiative for Safe AI (LISA) is a dedicated AI safety center and ecosystem located in central London. Its core mission is to foster an environment conducive to safely navigating the challenges and opportunities presented by transformative AI. LISA achieves this by enhancing the capacity, capability, and cooperation within the AI safety field. The organization operates a world-class hub that brings together various stakeholders, including AI safety organizations, training programs, independent researchers, and policymakers. Backed by leading funders and led by a team with deep AI safety expertise and operational experience, LISA provides workspace, strategic advisory, operational support, and community events to mobilize talent, strengthen organizations, and convene the broader AI safety ecosystem.

DRLib is a concise deep reinforcement learning library designed to integrate almost all off-policy RL algorithms with Hindsight Experience Replay (HER) and Prioritized Experience Replay (PER). Built upon OpenAI's spinningup, it offers implementations in both TensorFlow and PyTorch. The library simplifies the original spinningup code by removing multi-process and experimental grid wrappers for easier application and debugging. It also incorporates the D2SR method for efficient reward function design. DRLib is particularly well-suited for robotics-related tasks, providing encapsulated DDPG, TD3, and SAC algorithms with GPU support for PyTorch.

Dface is an open-source deep learning project focused on face detection and recognition, built with PyTorch. It offers comprehensive functionalities including face tracking, detection, recognition, and anti-spoofing. The tool leverages PyTorch's reverse-mode auto-differentiation for dynamic and easily reviewable code, and supports GPU acceleration with NVIDIA CUDA for real-time performance. Dface's architecture is inspired by research papers like Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks and FaceNet. It provides tutorials for training both detection (MTCNN) and recognition models, and includes pre-trained models for immediate testing. The project encourages community contributions and offers support channels for users.

DocRes is an open-source AI tool designed to unify various document image restoration tasks into a single generalist model. It addresses common issues found in scanned or photographed documents, such as dewarping distorted pages, removing shadows, enhancing appearance, deblurring text, and binarizing images for better readability. The tool provides a comprehensive solution for improving the clarity and quality of document images, making them more suitable for further processing like OCR. DocRes is implemented in Python and available on GitHub, offering scripts for inference, evaluation, and training, along with pre-trained model weights for immediate use.

LegitGrails offers a luxury authentication service for designer items, leveraging over six years of expert human authentication combined with advanced AI technology. The platform verifies the authenticity of handbags, sneakers, streetwear, and other luxury goods, providing results in 30 minutes or less with over 99.3% accuracy. It caters to both individual consumers and businesses, including consignment stores, online marketplaces, and pawn shops. Users upload photos of their item, and LegitGrails' team reviews them before issuing a certificate of authenticity. The service covers over 550 brands, from Hermès to Nike, and offers features like order tracking, cashback credits, and a refund guarantee.

FedLab is a highly customizable and flexible Federated Learning Framework built on PyTorch, designed to simplify research in this burgeoning area of machine learning. It provides essential modules for FL simulation, including communication, compression, model optimization, and data partition, allowing users to construct FL environments with custom components. FedLab supports various federated learning paradigms and algorithms, offering reproductions of baseline algorithms like FedAvg and FedProx. It also provides extensive data partitioning schemes for both IID and non-IID scenarios, along with visualization tools, making it easier for researchers to manage and experiment with diverse datasets.

Gollama is an open-source command-line tool designed for macOS and Linux users to efficiently manage their Ollama models. It offers a Text User Interface (TUI) that simplifies tasks such as listing, inspecting, deleting, copying, and pushing models. Users can interactively select models, sort them by various criteria like name, size, or modification date, and filter them. The application also supports editing Modelfiles, running and unloading models, and calculating approximate vRAM usage. A key feature is its comprehensive vRAM estimation, which helps users determine optimal quantization settings and context lengths for their hardware.

ShopPal AI is a U.S.-based company specializing in providing AI infrastructure integration, operational enablement, and go-to-market execution services exclusively for the fashion and apparel industry. They assist fashion companies in translating emerging AI technologies, such as shopping agents, visual AI, content AI, and marketing automation, into deployable, scalable, and revenue-generating business systems. ShopPal AI aims to elevate AI products beyond standard mobile internet applications by creating innovative and impactful solutions. Their services include AI system integration across e-commerce, content, marketing, and data platforms, end-to-end implementation, workflow design, and post-launch optimization. They also offer U.S. market entry support and AI-enabled content production and marketing campaigns.

FreeNoise is presented as a Hugging Face Space, developed by MoonQiu (Haonan Qiu). While the live website indicates the space is currently sleeping due to inactivity, its presence on Hugging Face suggests it's an AI tool primarily for experimentation and potentially noise generation. Tools hosted on Hugging Face Spaces are often used for showcasing machine learning models, research, and educational purposes, allowing users to interact with AI applications directly in a web browser. The tool's status as a 'sleeping' space implies it's a project that can be restarted and utilized for its intended functions, likely within the domain of AI research or development.

Meteroid is an open-source monetization platform designed for SaaS companies to launch, test, and scale their business models. It offers comprehensive features for managing subscriptions, invoicing, metering, and quotes, enabling businesses to monetize new features quickly and accurately. The platform helps streamline revenue from quote to cash, empowering sales teams with custom pricing options and eliminating manual work. Built with Rust for performance and reliability, Meteroid processes millions of usage events in real-time. Its API-first architecture ensures seamless integration into existing tech stacks, and being open-source, it provides flexibility and future-proofing.

Roton Consultancies Private Limited is an export consulting firm dedicated to helping Indian MSMEs scale globally. They offer comprehensive services including market entry and strategy development, buyer access and distributor mapping, and compliance and export enablement. Roton assists organizations in identifying target markets, building go-to-market plans, and connecting with verified global partners. Their expertise covers industry-specific standards and certifications (e.g., GOTS, REACH, HACCP) and they provide support for trade fair participation and buyer meetings. Roton focuses on delivering measurable outcomes within weeks, offering practical solutions like verified distributor lists, country-wise compliance checklists, and booked buyer introductions across sectors such as Textiles, Gems & Jewelry, and Chemicals.

LINE (Large-scale Information Network Embedding) is a toolkit designed for embedding very large-scale information networks. It supports various network types, including directed, undirected, binary, and weighted edges. The tool is highly efficient, capable of embedding networks with millions of vertices and billions of edges on a single machine within a few hours. While this specific GitHub repository is no longer maintained, it provides the source code and documentation for the original LINE model. Users can compile and run the tool on both Windows and Linux, with external packages like BOOST (for Windows) and GSL (for Linux) required for random number generation in the edge-sampling algorithm. The toolkit includes functionalities for reconstructing sparse networks, normalizing embeddings, and concatenating first-order and second-order embeddings.

Proof Crater is a privacy-first commitment engine designed for blockchain data, allowing teams to anchor complex datasets on-chain and verify them later using zero-knowledge proofs and cryptographic evidence. Its flagship ZK Vault functionality facilitates fully anonymous, non-custodial asset distribution, ideal for private payrolls or airdrops. Recipients can claim funds to any clean address via a relayer, with proofs generated locally to ensure privacy. Beyond payments, it serves as a robust, independently verifiable snapshot registry, capturing historical on-chain state for audits, governance, and compliance. Proof Crater is chain-agnostic and prioritizes privacy and scrutiny over speed, providing a foundation for long-lived, auditable obligations.

Ali Vilab In Context LoRA is an AI tool hosted on Hugging Face Spaces, designed to generate text responses from user-provided prompts. This application allows users to input a text prompt, and the underlying model will process it to produce a relevant text output. While the tool's specific focus on "In Context LoRA" suggests an emphasis on leveraging Low-Rank Adaptation within AI models for contextual text generation, the current status indicates it is inactive. It provides a platform for experimenting with text generation capabilities, likely targeting researchers or developers interested in AI model behavior and text-based applications.

Strantin is a comprehensive and user-friendly enterprise video surveillance software designed for monitoring, reviewing, and streamlining images from security systems. It integrates powerful computing with digital video technologies to create highly functional surveillance solutions. The software is constantly enhanced to allow clients to build complex surveillance systems tailored to their evolving needs. Strantin offers features like Op Center™, Scene Enhancer, System Diagnostics, Site Mapping, Chronology, Storage, Dispatch Command, and Guided Setup. It also provides NVR hardware series (Core, Professional, Lineage, Enterprise) and various deployment models, making it suitable for multiple-site environments where video traffic is directed to centralized servers.

Pixee AI functions as an Agentic AppSec platform, designed to triage and automatically fix software vulnerabilities. It boasts a 98% noise reduction rate by eliminating false positives through exploitability analysis, ensuring only real risks are addressed. The platform generates context-aware fixes that match coding conventions and security rules, achieving a 76% developer merge rate. Pixee learns continuously from team actions, adapting to specific coding conventions, policies, and preferences to become an autonomous product security engineer. It integrates deep codebase analysis, security policy integration, and execution path tracing to understand the real attack surface and prioritize risks effectively. This solution helps enterprises clear security backlogs, scale AppSec capacity, and ensure compliance by automating vulnerability resolution.

StemGNN is an open-source implementation of a Spectral Temporal Graph Neural Network designed for multivariate time-series forecasting. Developed by Microsoft, this tool is implemented in Python and optimized for Ubuntu 18.04.2 LTS. It allows users to train and evaluate models on diverse datasets, including traffic flow (PEMS03, PEMS04, PEMS07, PEMS08, METR-LA, PEMS-BAY), solar power, electricity consumption, ECG data, and COVID-19 case numbers. The repository provides detailed instructions for setting up a virtual environment, installing dependencies, and running training and evaluation procedures with customizable parameters such as window size, prediction horizon, and normalization methods. It also includes pre-cleaned ECG data for convenience.

StringSifter is an open-source machine learning tool developed by Mandiant that automatically ranks strings based on their relevance for malware analysis. This capability significantly speeds up the analysis process by highlighting the most critical strings, allowing analysts to focus on high-priority indicators. The tool can process strings extracted from various sources, including binaries, memory dumps, or sandbox runs, and is compatible with outputs from other tools like FLOSS. It provides command-line utilities for both string extraction (flarestrings) and ranking (rank_strings), supporting batch processing and filtering options. StringSifter requires Python 3.9 or newer and can be installed via pip or run within a Docker container, making it accessible for developers and security researchers.

svox2 is an open-source project that implements Plenoxels, a technique for representing radiance fields without relying on neural networks. This approach offers an alternative to traditional neural radiance fields for 3D scene representation and rendering. The tool is designed for researchers and developers in computer vision and 3D graphics, providing official optimization code and support for various dataset formats like NeRF-Blender, LLFF, NSVF, and CO3D. It includes scripts for data processing, optimization, and rendering, with options for single scene training or parallel task execution. While primarily tested on Linux, it offers a robust framework for experimenting with radiance fields.

Tagger is an open-source implementation of a Named Entity Recognizer (NER) that delivers state-of-the-art performance across four CoNLL datasets: English, Spanish, German, and Dutch. A key differentiator is its ability to achieve this high level of accuracy without relying on any language-specific knowledge or external resources like gazetteers. The tool provides a straightforward command-line interface for tagging sentences using pre-trained models or for training custom models with user-provided datasets. It requires Python 2.7 with Numpy and Theano installed, making it accessible for researchers and developers familiar with these environments. The project is hosted on GitHub under an Apache-2.0 license, encouraging community contributions and further development.

Tianshou (天授) is a reinforcement learning (RL) library built on pure PyTorch and Gymnasium, designed for both RL researchers and practitioners. It offers modular low-level interfaces for algorithm developers, emphasizing flexibility, hackability, and type-safety, alongside convenient high-level interfaces for applying RL to custom environments. The library supports a wide range of RL paradigms, including online (on- and off-policy) and offline RL, with experimental support for multi-agent RL (MARL) and model-based RL. Tianshou stands out for its high-performance, modularized framework, user-friendly interfaces, and generality, enabling concise implementations without sacrificing flexibility. It includes implementations of numerous state-of-the-art algorithms like DQN, PPO, SAC, and various offline RL methods, and supports features such as vectorized environments, recurrent state representations, and multi-GPU training.