Projects

### Overview: Alts is a local and offline voice assistant that uses speech recognition and text-to-speech capabilities. It listens to your voice, thinks about your queries, and responds with synthesized speech, all without leaving your environment. ### Key Features: - **Voice Recognition:** Records your voice and transcribes it using OpenAI's Whisper model. - **Text-to-Speech:** Converts your queries into spoken words using Coqui-TTS. - **Local Operation:** Runs entirely within your local environment, ensuring privacy and no data leaving your system. - **Configurable Hotkey:** Allows you to customize the hotkey combination for starting and stopping the assistant. - **Private Recording:** All recordings are deleted as soon as they are used, ensuring your data remains secure. ### Tech Stack: Python ### Why it Matters: Alts is a free, local, and offline voice assistant that can be used for quick responses and reminders. It's perfect for those who prefer to keep their conversations private and avoid the internet.

### Overview: The GPT Image 2 (official) API (`gpt-image-2-official`) provides access to the OpenAI-compatible APIMart endpoint, allowing developers to generate high-quality images using the GPT-2 model. This API is priced on a token-billed model, with flexible pricing tiers and a simple API structure. ### Key Features: - **One Key, Entire Catalog**: The same `https://api.apimart.ai/v1` base URL and `Authorization` header reach GPT Image 2 (official) and 300+ other image, video, and language models. - **$1 Minimum, Pay as You Go**: No subscription required; pay only for the tokens used. - **Charge Back to Response**: Every completed task reports the amount billed, and tasks marked as failed are refunded. - **Async by Design**: Submit, take the `task_id`, and poll `GET /v1/tasks/{id}` for results, with batch processing and retries as normal queue work. ### Tech Stack: - Python ### Why It Matters: - **One Key, Entire Catalog**: The same base URL and `Authorization` header reach all models, making it easy to switch between different models. - **$1 Minimum, Pay as You Go**: No subscription required, and pay only for the tokens used, making it flexible and cost-effective. - **Charge Back to Response**: Every completed task reports the amount billed, ensuring transparency and fairness. - **Async by Design**: Submit, take the `task_id`, and poll for results

Overview: This project provides a GPT Image 2 (official) API endpoint that allows developers to access the OpenAI-compatible APIMart API, enabling them to generate images using the GPT-2 model. The API supports various parameters and pricing models, making it easy to integrate into their projects. Key features: - Access to the GPT-2 model via the APIMart endpoint - Support for different image resolutions and output types - Pricing based on token usage, with a minimum charge of $1 - Billing is applied directly to the response, ensuring transparency Tech stack: Python Why it matters: This API simplifies the integration of GPT-2 into projects, providing a consistent and efficient way to generate images without the need for complex integration or setup. The token-based pricing model ensures cost transparency, making it suitable for both small and large-scale projects. Repository: apimartlzras0rk/gpt-image-2-official-api-ko Primary language: Python GitHub stars: 81 GitHub description: GPT Image 2 (official) API (`gpt-image-2-official`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

### Overview: InferenceDB is a cloud-native, model-serving tool designed to stream real-time inferences of ML models from production environments into a data lake. It supports multiple model serving frameworks and integrates with various cloud infrastructures, enabling efficient model retraining, data drift monitoring, and incident investigation. ### Key Features: - **Cloud Native**: Runs on Kubernetes and supports any cloud infrastructure. - **Model Serving Integrations**: Connects to ML model serving tools like KServe. - **Extensible**: Allows adding custom model serving frameworks and database destinations. - **Horizontally Scalable**: Supports adding more workers to handle more models and traffic. - **Python Ecosystem**: Written in Python using Faust, allowing for data transformations with Numpy, Pandas, etc. ### Tech Stack: InferenceDB is built using Python, Faust, and Kafka. It leverages Faust for model serving integrations and Kafka for event streaming. ### Why It Matters: InferenceDB simplifies the process of streaming model inferences to a data lake, making it easier to monitor and analyze model performance in real-time. This tool is particularly useful for AI incident investigation, model retraining, and data drift monitoring in production environments. ### GitHub Stars: The repository has been starred by 82 developers, indicating high interest and engagement in the project. ### Installation: To install InferenceDB, you need a Kafka cluster with Schema Registry and Kafka Connect. The Helm chart can be used to install InferenceDB on Kubernetes. ### Usage: InferenceDB is designed to

### Overview: This repository is a curated list of resources, tools, and libraries for the Stable Diffusion project, focusing on modern diffusion image models. It includes UIs, libraries, training methods, and more, aiming to streamline and enhance the development and utilization of Stable Diffusion. ### Key Features: - **ComfyUI**: A modular node-based GUI for diffusion models. - **Diffusers**: A state-of-the-art library for diffusion models in PyTorch. - **kohya_ss**: A GUI for training LoRA and DreamBooth models. - **Stable Diffusion web UI**: A widely used Gradio-based Stable Diffusion web UI. - **automatic**: An opinionated fork of Stable Diffusion WebUI. - **Fooocus**: A minimalist, user-friendly interface. - **InvokeAI**: A creative engine for professionals and enthusiasts. - **SwarmUI**: A high-performance web UI for image and video diffusion models. - **HEISS UI**: A local front end for ComfyUI with 28 built-in models and custom workflow support. - **stable-diffusion-webui-forge**: An optimized A1111-based WebUI with improved VRAM management. - **sd-webui-forge-classic**: A FLUX-based Neo fork with modern model support. - **ComfyUI-KJNodes**: Utility nodes and model optimization for ComfyUI. - **ComfyUI-LayerStyle**: Photoshop-style layer, mask, and compositing nodes for ComfyUI. - **ComfyUI-Custom-

Overview: GPT Image 2.5 API offers high-resolution images generated by the GPT model, with prices starting at $0.0085 per unit. It supports flare@1K and sunburst@1K, with flare@2K priced at $0.014. Key features: - High-resolution images - Flare@1K and Sunburst@1K - Pricing starting at $0.0085 per unit - Support for flare@2K at $0.014 - OpenAI-compatible API gateway Tech stack: Shell Why it matters: This API provides high-quality images generated by the GPT model, making it suitable for various creative and design projects without the need for expensive professional models.

Overview: CAPPr is a Python library that allows developers to integrate the Completion After Prompt Probability (CAPPr) functionality into their projects. This feature enables LLMs to make choices based on a given prompt, or compute the probability of a completion given a prompt. Key features: - Use a GGUF model for high-performance inference - Use a Hugging Face transformers model for compatibility with existing models - Cache instructions to save time and improve performance Tech stack: Python Why it matters: CAPPr provides a powerful tool for developers to enhance their LLMs with decision-making capabilities based on prompts, making them more versatile and useful in various applications.

### Overview: WordStylist is an official PyTorch implementation of the "WordStylist: Styled Verbatim Handwritten Text Generation with Latent Diffusion Models" paper presented at ICDAR 2023. This project aims to generate realistic word image samples from different writer styles using latent diffusion models without the need for adversarial training, writer recognition, or text recognition. The system is evaluated using the Fréchet Inception Distance, writer recognition accuracy, and writer retrieval, demonstrating aesthetically pleasing results that enhance text recognition performance. ### Key Features: - **Latent Diffusion Models**: Utilizes latent diffusion models for styled text-to-image generation. - **Class Index Styles**: Generates samples from different writer styles using class index styles. - **No Adversarial Training**: Reduces the need for adversarial training, writer recognition, and text recognition. - **Performance Metrics**: Measured using Fréchet Inception Distance, writer recognition accuracy, and writer retrieval scores. ### Tech Stack: - **Python**: The primary programming language used for this project. ### Why It Matters: WordStylist offers a novel approach to styled text-to-image generation, making it a valuable tool for enhancing text recognition and improving the aesthetic quality of generated images. Its ability to generate realistic samples from different writer styles without the need for complex training methods makes it particularly useful in various applications, including data augmentation and document image processing. ### Repository: - **Repository**: `koninik/WordStylist` - **GitHub Stars**: 82 - **GitHub Description

### Overview: This repository serves as a comprehensive survey of techniques, applications, and future directions in generative AI for character animation. It aims to provide researchers and practitioners with a detailed overview of the latest advancements in the field, including computer graphics models, vision-based models, language models, temporal sequence modeling, speech models, and additional generative models. ### Key features: - 📖 Papers, datasets, and resources related to generative AI for character animation. - Continuous updates to keep the repository relevant and up-to-date. - Cited papers and datasets for further reading and contributions. ### Tech stack: - Python, Jupyter notebooks, and other programming languages used for data analysis and visualization. ### Why it matters: This repository is crucial for researchers and practitioners in the field of generative AI for character animation. It serves as a valuable resource for staying informed about the latest techniques, applications, and future directions in the field. The comprehensive nature of the repository ensures that users can easily find and understand the most relevant information for their projects.

Overview: Parea AI provides a Python SDK for evaluating and monitoring LLM-powered applications. It includes quickstarts for testing, logging, and deploying prompts. Key features: - Quickstarts for evaluating and testing LLM apps - Automatic logging of LLM calls to OpenAI - Instrumentation for logging and observability - Support for deploying prompts for collaboration between engineers and subject matter experts Tech stack: Python Why it matters: Parea AI simplifies the process of evaluating and monitoring AI applications, making it easier for developers to experiment and improve their models without the need for extensive setup or configuration.

Overview: Sytody is a Flutter "speech to todo" app example that demonstrates how to integrate speech recognition and speech-to-text functionality into a Flutter application. Key features: - Speech recognition - Speech-to-text functionality - Flutter integration Tech stack: Dart Why it matters: Sytody showcases the integration of speech recognition and speech-to-text capabilities into a Flutter application, providing a practical example for developers looking to enhance their Flutter projects with natural language processing. Repository: rxlabz/sytody Primary language: Dart GitHub stars: 82 GitHub description: a Flutter "speech to todo" app example README excerpt: # Sytôdy, a Flutter "speech to text" todo app POC :warning: iOS10(Swift) & Android [](https://youtu.be/7MGuNZfgGWw) ## Usage Install [flutter](http://flutter.io) ```bash cd sytody flutter run ``` :tv: [Video demo](https://youtu.be/7MGuNZfgGWw) ## How it works **TL;DR** The flutter app opens a channel on the host platform. It uses it: - to ask for speech recognition and microphone usage. The permission is asked on the first application launch - to start, cancel and stop the speech recognition - to listen to the recognition result **Flutter, native APIs and plugins** - [1. Sytody](https://medium.com/@rxlabz/flutter-native

Overview: This project, "Impossible Videos," aims to explore and evaluate advanced video generation and understanding models by generating and understanding impossible video scenes that defy physical, biological, geographical, or social laws. It introduces a novel benchmark, IPV-Bench, to assess models' abilities in following prompts and understanding impossible videos, with a focus on temporal dynamics and world knowledge. Key features: - Introduction of a comprehensive taxonomy for impossible video scenes. - Construction of a prompt suite based on the taxonomy to evaluate video generation models. - Curated video benchmark to assess Video-LLMs on their ability to understand impossible videos, requiring reasoning on temporal dynamics and world knowledge. Tech stack: The project utilizes Python for implementation, with contributions from multiple researchers and developers. Why it matters: This project is significant for advancing video research by addressing key questions in video generation and understanding, such as whether today's models can effectively generate impossible video content and whether they can understand impossible videos well.

Overview: This project is an official implementation of the ECCV 2024 paper "Beyond the Contact: Discovering Comprehensive Affordance for 3D Objects from Pre-trained 2D Diffusion Models". It provides a comprehensive dataset and detailed instructions for generating 2D HOI images, 3D HOI samples, and learning ComA using pre-trained 2D diffusion models. Key features: - Pre-trained 2D diffusion models for generating 2D HOI images and 3D HOI samples. - Comprehensive dataset of 83 object meshes from SketchFab, including various interaction types. - Detailed documentation and instructions for setting up the environment and running the pipeline. - Visualization tools for human contact, object contact, and human orientational tendency using Blender and Mayavi. Tech stack: Python. Why it matters: This project is significant for researchers and developers interested in 3D object affordance and interaction analysis, offering a practical tool for training and evaluating models on complex 3D object interactions using pre-trained 2D diffusion models. Repository: snuvclab/coma GitHub stars: 82 GitHub description: Official Repository for ECCV 2024 paper "Beyond the Contact: Discovering Comprehensive Affordance for 3D Objects from Pre-trained 2D Diffusion Models".

### Overview: This repository is a curated list of papers on contextualizing End-to-End Automatic Speech Recognition (ASR) outputs. The purpose is to improve the accuracy and relevance of transcription by biasing the results towards specific tokens, such as proper nouns, rare words, or jargon, which are likely to be produced given the context of an audio signal. ### Key Features: - **Deep Contextualization:** - **Contextual LAS (CLAS):** Deep context-aware speech recognition. - **Contextual Transducer (RNNTs):** End-to-end speech recognition with contextualized transducers. - **2021:** - **Tree-constrained Pointer Generator:** Improves contextual speech recognition. - **Fast Contextual Adaptation:** Enhances personalized speech recognition. - **2022:** - **Improving End-to-End Contextual Speech Recognition:** Enhances contextual speech recognition. - **End-to-end contextual ASR based on posterior distribution adaptation:** Enhances hybrid CTC/attention system. - **Towards Contextual Spelling Correction:** Enhances customization of speech recognition systems. ### Tech Stack: The repository uses Python for data processing and analysis, and it is maintained using GitHub. ### Why It Matters: Correct transcription of these tokens can significantly improve the value of the output, while incorrect transcription might otherwise be likely. By contextualizing ASR outputs, the system can better align with the intended meaning and context of the audio, leading to more accurate and relevant transcriptions.

Overview: ASK MIRAGE is an offline AI-powered chatbot that allows users to upload PDFs and ask natural language questions about the content. Built with LangChain, FAISS, and LLaMA, it delivers fast, context-specific responses without requiring internet connectivity. Key features: - Fully Offline Access - PDF Upload & Parsing - Context-Aware Responses - Natural Language Queries - Vector-Based Search - Modular Design - Privacy First - Quick Response Time Tech stack: Python, Flask, LangChain, LLaMA, FAISS, PyPDF, HTML, CSS, JavaScript Why it matters: Ideal for students and professionals working with large text documents, ASK MIRAGE provides fast, context-aware responses without the need for an internet connection.

Overview: Noema is a declarative framework designed to control and interpret Large Language Models (LLMs) in Python, providing developers with a seamless integration between Python and LLM generations. Key features: - Seamless integration between Python and LLM generations. - Control over the model's path and behavior. - Built on top of llama.cpp and guidance-ai. Tech stack: Python Why it matters: Noema simplifies the interaction with LLMs, allowing developers to interpret and control them effectively, rather than relying solely on their outputs as a source of truth. This approach enhances the interpretability and flexibility of LLMs, making them more suitable for applications that require human-like reasoning and decision-making.

Overview: This project integrates Neo4j with AI models to generate optimized Cypher queries from natural language prompts, demonstrating the power of AI in enhancing Neo4j's capabilities for AI-driven RAG systems. Key features: - AI-powered natural language to Cypher query conversion - Neo4j integration with vector search capabilities - RAG (Retrieval-Augmented Generation) example included - Database seeding for reproducible testing - Dockerized Neo4j instance Tech stack: JavaScript Why it matters: This project showcases the potential of combining Neo4j with AI models to create more efficient and intelligent querying systems, particularly in the realm of AI-driven RAG applications. Repository: ErickWendel/neo4j-ai-experiments GitHub stars: 81 GitHub description: Examples of my tutorial on how to use Neo4j for empowering AI RAG systems

### Overview: This repository contains design patterns for various multi-agent frameworks, including Autogen, Langraph, Taskweaver, and CrewAI. The patterns aim to provide a standardized approach for building autonomous AI agents and related systems. ### Key Features: - **Agent Architecture**: A comprehensive overview of agent design patterns. - **Planner Component**: Detailed illustration of the planner component in agent architectures. - **Taskweaver Design**: Visual representation of Taskweaver's architecture. - **Autogen Agent Chat**: An example of an agent chat using the Autogen framework. - **Multi-Agent Conversation**: A demonstration of multi-agent conversations. - **Hugging GPT Agent**: A showcase of an agent using the Hugging GPT framework. ### Tech Stack: Jupyter Notebook ### Why it Matters: These design patterns provide a standardized and systematic approach to building multi-agent systems, ensuring consistency and efficiency across different frameworks.

Overview: This repository contains open-source scripts and pipelines developed by the MICA lab, including tools for microstructure profile covariance analysis, gradients, BigBrain profiles, BrainSpace, diffusion map embedding, a moment of change, SurfStat, and more. These resources are useful for analyzing microstructural gradients, neuroimaging, and connectome data. Key features: - MPC: Microstructure profile covariance analysis tool - gradients_volumetric: Gradients from Paquola et al. (2019) - BigBrain: Histological profiles and statistical moments - BrainSpace: Macroscale gradient mapping and analysis - diffusion_map_embedding: Diffusion map embedding technique - a_moment_of_change: Companion to the "Shifts in myeloarchitecture characterise adolescent development of cortical gradients" preprint - SurfStat: Surface- and volume-based statistical analysis - mica_powertools: Scripts from the MICA lab - dcm_sorter: Sorts DICOM directories - autism_hierarchy: Videos of stepwise functional connectivity analysis Tech stack: Jupyter Notebook Why it matters: These tools and resources are essential for researchers and clinicians working in the field of multimodal imaging and connectomics, providing a comprehensive set of tools for analyzing neuroimaging and connectome data.

### Overview: DJZ-Nodes is a comprehensive collection of custom nodes for ComfyUI, featuring over 75 specialized nodes that revolutionize advanced image/video processing, audio manipulation, batch operations, prompt engineering, and workflow optimization. It includes professional video & film effects, advanced video processing, creative tools suite, intelligent prompt engineering, and workflow optimization features. ### Key Features: - **Professional Video & Film Effects**: Cinematic lens effects, vintage & creative effects, datamoshing, interlaced video simulation, bit-depth manipulation, and more. - **Advanced Video Processing**: Eight generations of datamoshing nodes, multiple interlaced video simulation methods, frame management, transition effects, and bit-depth manipulation. - **Creative Tools Suite**: Fractal generation, parametric 3D mesh creation, geometric patterns, three-tone styling, and advanced image processing. - **Intelligent Prompt Engineering**: Advanced prompt manipulation, Zenkai system, dynamic wildcards, duplicate removal, and smart text processing. - **Workflow Optimization**: Comprehensive batch operations, file management, and dataset tools. - **Audio Integration**: Professional audio mixing and merging capabilities. ### Tech Stack: Python ### Why it Matters: DJZ-Nodes offers a complete set of custom nodes for ComfyUI, providing advanced tools for creative media workflows, making it a valuable asset for artists and developers alike.