Projects

### Overview: A11 is an open-source runtime designed to simplify the development of AI applications, allowing developers to build and deploy models, tools, and interfaces with ease. ### Key features: - **Named Streams**: Instead of using mixed provider events, A11 uses named streams to read model text, reasoning, tool activity, progress, and durable interaction state from separate ports. - **Same Action Contract Locally and Remotely**: The same action contract can be used both locally and remotely, enabling seamless integration across different languages or processes. - **Deterministic Runtime Composition with Flow**: A11 allows for the composition of deterministic runtime with Flow, enabling runtime composition with defined concurrency, draining, deadlines, cancellation, and error propagation. ### Tech stack: C++ ### Why it matters: A11 simplifies the development of AI applications by providing a unified runtime that connects models, tools, and interfaces, making it easier for developers to build and deploy AI applications without the complexity of low-level programming.

### Overview: ReelTek Vision System ReelTek is a real-time camera analysis platform that utilizes local vision-language models to provide insights from your webcam feed. With `SmolVLM` and `TTS`, ReelTek offers real-time video analysis and object detection, integrated with a modern, user-friendly web interface. ### Key Features: - Real-time video analysis and object detection using webcam feed - Integration with lightweight vision-language models for efficient processing - Customizable analysis instructions for tailored output - System metrics monitoring for performance optimization - **Speech synthesis support** - Hear the AI's analysis spoken aloud - Browser-based TTS with multiple voice options - Optional Python TTS server for enhanced voice quality ### Tech Stack: HTML ### Why it Matters: ReelTek Vision System is designed to provide real-time insights into your camera feed, making it an essential tool for anyone looking to monitor their environment or analyze video content in real-time. The combination of `llama.cpp` and `SmolVLM` ensures efficient processing and real-time analysis, making it a powerful addition to any developer's toolkit. Repository: iBz-04/reeltek Primary language: HTML GitHub stars: 10 GitHub description: A small VLM that sees everything

Overview: Toolkit Manager is a production-ready platform for deploying OpenAI-compatible LLM inference on Intel® Xeon® silicon, including secure Keycloak authentication and built-in observability. It simplifies the process of deploying AI on Intel® Xeon® processors with a single installer script and customizable Kubernetes-based infrastructure. Key features: - Secure Keycloak authentication - Built-in observability - Production-ready AI stack - One-command installer for enterprise generative AI Tech stack: Shell Why it matters: Toolkit Manager streamlines the deployment of AI on Intel® Xeon® silicon, making it easy for teams to connect LLM inference, RAG, and agentic workflows, and secure their infrastructure with a unified AI gateway.

### Overview: Qwen Image 2.1 WebGPU is a pure Rust implementation of the Qwen Image 2.1 model, running entirely in the browser using WebGPU. This model is optimized for low VRAM usage and high performance, making it a valuable tool for text-to-image generation. ### Key Features: - **Browser-Native Inference:** Runs on any GPU via the `wgpu` backend, featuring full WebGPU support. - **Low VRAM Footprint:** Implements Disk Layer Streaming to fetch and unload transformer blocks directly from the browser's Origin Private File System (OPFS) on demand. - **Optimized Compute:** Uses Flash Attention within the Diffusion Transformer (DiT) and Text encoder's Transformer to prevent materializing full N x N matrices, drastically saving memory. - **Quantization:** Leverages int8 block quantized transformer blocks for high-performance execution. ### Tech Stack: Rust ### Why it Matters: This model holds the Rank 1 spot on the Text-to-Image Arena in the Open Source category, making it a highly valuable tool for text-to-image generation.

Overview: This repository contains Kling 3.0 prompts, including 8 tested recipes with example renders, aspect ratios, the model ID (`kling-v3`), and the cost per image. It's a resource for developers looking to integrate Kling 3.0 into their projects. Key features: - 8 Kling 3.0 prompts with example renders, aspect ratios, and pricing - Model ID: `kling-v3` - Effective price: $0.0672 per second Tech stack: Python Why it matters: Kling 3.0 is a powerful image generation model that can be easily integrated into applications using the provided API key and base URL. It offers a cost-effective alternative to traditional image generation methods, making it ideal for developers seeking to streamline their workflows without the need for a monthly subscription.

### Overview: Qwen Image 2.0 prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`qwen-image-2.0`), and the cost per image. ### Key features: - **Model ID:** `qwen-image-2.0` - **Aspect Ratios:** 1:1, 4:3, 8:5, 3:4, 16:9 - **Prompt Structure:** - **Subject:** Concrete object or person - **Light:** Direction + quality - **Framing:** Ratio + lens or distance - **Setting:** Surface, backdrop, environment - **Finish:** Render or film character ### Tech Stack: Python ### Why it Matters: - **$0.02 per image** on the APIMart route — a unique endpoint compatible with OpenAI, available without a monthly plan. - **No subscription or prepaid plans** for scaling: recarge from $1 and charge per call. - **Chargeback on completion:** All tasks are billed upon completion; failed tasks are refunded. - **Asynchronous design:** Send tasks, retrieve `task_id`, and check `GET /v1/tasks/{id}` for retries and batching. ### Eight Prompts: - **Product / e-commerce:** A frosted glass perfume bottle on a wet black stone slab. - **Food photography:** A spicy miso ramen bowl with soft boiled egg, nori, and scallions. - **C

### Overview: Seedance 2.0 Fast Prompts: 8 Tested Recipes with Outputs <!-- conv-kit:v1 --> <p align="center"> <img src="assets/badges/price.svg" alt="observed unit price"> <img src="assets/badges/billing.svg" alt="billing model"> <img src="assets/badges/compat.svg" alt="OpenAI-compatible endpoint"> </p> <p align="center"> <img src="assets/01-product-ad-ceramic-mug-thumb.jpg" width="820" alt="Sample video frame generated through APIMart"> </p> > **from $0.0213 per second** on the APIMart route (billed by resolution tier) — un seul endpoint compatible OpenAI sur `https://api.apimart.ai/v1`, sans abonnement mensuel. *(relevé le 2026-10-12)* **[Get an API key](https://go.apimart.ai/k-4778df)** · **[Live pricing](https://go.apimart.ai/k-af560e)** · **[Model page](https://go.apimart.ai/k-a2f12d)** · [⚡ 60-second quickstart](#quickstart) **Pourquoi les équipes passent par APIMart pour Seedance 2.0 Fast (`doubao-seedance-2.0-fast`)** - **Une

Overview: This repository contains 8 tested Nano Banana Pro prompts with example renders, aspect ratios, the model ID (`gemini-3-pro-image-preview`), and the cost per image. Key features: - 8 Nano Banana Pro prompts - Example renders, aspect ratios, model ID, and cost per image - $0.03 per image (4K $0.04) - OpenAI-compatible endpoint `https://api.apimart.ai/v1` - No subscription required, pay-as-you-go pricing - Real-time cost reporting, automatic refunds for failed tasks Tech stack: Python Why it matters: This repository provides a concise guide for developers to create effective image prompts using Nano Banana Pro, ensuring high-quality results without the need for complex setup or prepayment.

### Overview: Kling 3.0 Turbo Prompts: 8 Tested Recipes With Outputs <!-- conv-kit:v1 --> <p align="center"> <img src="assets/badges/price.svg" alt="observed unit price"> <img src="assets/badges/billing.svg" alt="billing model"> <img src="assets/badges/compat.svg" alt="OpenAI-compatible endpoint"> </p> <p align="center"> <img src="assets/01-product-ad-ceramic-mug-thumb.jpg" width="820" alt="Sample video frame generated through APIMart"> </p> > **from $0.1144 per second** on the APIMart route — um único endpoint compatível com OpenAI em `https://api.apimart.ai/v1`, sem plano mensal. *(observado em 2026-10-12)* **[Get an API key](https://go.apimart.ai/k-195cb2)** · **[Live pricing](https://go.apimart.ai/k-ad43f3)** · **[Model page](https://go.apimart.ai/k-5c7409)** · [⚡ 60-second quickstart](#quickstart) **Por que os times usam Kling 3.0 Turbo (`kling-3.0-turbo`) via APIMart** - **Uma chave, o catálogo inteiro.** A mesma base

### Overview: FLUX 2 Max Prompts: 8 Tested Recipes with Outputs ### Key Features: - **Model ID:** `flux-2-max` - **Effective Price:** $0.056 per image (1MP $0.056, 2MP $0.08, 3MP $0.104) - **Aspect Ratios:** 1:1, 4:3, 8:5, 8:10, 16:9, 16:5, 2:3, 3:2 - **Cost on This Route:** $0.056 per image ### Tech Stack: Python ### Why It Matters: FLUX 2 Max (`flux-2-max`) is a powerful tool for generating high-quality images, videos, and text prompts. With its optimized pricing and flexible aspect ratios, it's ideal for developers looking to streamline their workflows and produce professional-grade results without the hassle of setup fees or long-term commitments.

Overview: Vidu Q3 API (`viduq3`) is a powerful OpenAI-compatible API endpoint that allows developers to access a wide range of image, video, and language models. It offers a single API key for all models, eliminating the need for separate API keys and monthly fees. Key features: - One key for all models, including 300+ image, video, and language models. - Pricing is based on generated seconds, with no need for prepayment. - Summaries are returned in the response, making it easy to track progress. - Tasks are processed asynchronously, ensuring efficient handling of tasks. Tech stack: Python Why it matters: Vidu Q3 API simplifies API key management, eliminates the need for separate API keys and monthly fees, and provides a straightforward way to integrate with OpenAI-compatible endpoints. Repository: j5vhnckx497ss9/viduq3-api-ru Primary language: Python GitHub stars: 11 GitHub description: Vidu Q3 API (`viduq3`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

### Overview: This project focuses on developing a generative text-to-image model that can predict the text prompts used to generate images from Stable Diffusion 2.0. The model utilizes ensemble techniques combining Vision Transformer, CLIP Interrogator, and OFA fine-tuned for image captioning. ### Key Features: - Predicting text prompts from generated images - Utilizing Stable Diffusion 2.0 for image generation - Fine-tuning OFA for image captioning - Ensemble of Vision Transformer, CLIP Interrogator, and OFA models ### Tech Stack: - Jupyter Notebook ### Why it Matters: This project aims to understand the latent relationship between text prompts and generated images, providing insights into the reversibility of the latent relationship between text and images.

### Overview: 本仓库提供了一个Java AI框架的对比示例，通过对比Spring AI、LangChain4j和SAA三个框架，展示了它们在代码量、配置方式和运行表现上的差异。通过对六维维度的实测，对比了模型集成数量、向量库支持、MCP支持、多Agent编排、可观测性和中文社区的差异。 ### Key features: - **Spring AI 2.0**: 最小可运行的ChatClient实现，包含10行Demo。 - **LangChain4j 1.20**: 最小可运行的ChatService和@V模板变量。 - **SAA 1.0**: 最小可运行的百炼starter。 - **6维度实测脚本**: 包含模型集成、向量库、MCP、Agent、可观测性和社区的实测结果。 - **一键起pgvector、qdrant、redis8**: Docker-compose配置。 ### Tech stack: Java ### Why it matters: 通过对比这三大Java AI框架，帮助开发者了解它们在实际应用中的表现和适用场景，从而做出更合适的选择。

### Overview: This repository contains the codebase for a Retrieval-Augmented Generation (RAG) chatbot I built for my technical blog. The chatbot uses Langchain for the backend, LangServe for API endpoints, and Langfuse for monitoring. The chatbot is designed to answer questions about my blog and provides a conversational interface. ### Key Features: - **Backend**: Built with Langchain for RAG logic, FastAPI for API endpoints, Langfuse for monitoring, and Chroma as the Vector Database. - **Frontend**: Developed using NextJS and Chakra UI for the user interface, LangchainJS for interaction with backend APIs. - **Deployment**: Containerized using Docker and orchestrated with Docker Compose for lightweight deployment. ### Tech Stack: - **Backend**: Langchain, FastAPI, LangServe, Langfuse, Chroma, LangchainJS - **Frontend**: NextJS, Chakra UI, Docker, Docker Compose ### Why it Matters: This chatbot is a practical application of RAG technology, enhancing the user experience by providing accurate and conversational answers to questions about the blog. It showcases the integration of cutting-edge AI technologies in a real-world application, making it valuable for developers interested in AI and chatbot development.

Overview: This project provides a comprehensive API for generating images using the GPT-Image 2.5 model, compatible with OpenAI's API endpoint. Key features: - Access to the GPT-Image 2.5 model - Pricing based on the number of images generated - Support for different image resolutions - Asynchronous processing Tech stack: Python Why it matters: This API simplifies the process of generating high-quality images for developers and designers, making it easier to integrate into their workflows without the need for complex integration or setup. Repository: jfblmhen202p/gpt-image-2.5-api-es Primary language: Python GitHub stars: 11 GitHub description: ChatGPT Imágenes — GPT Image 2.5 API (`gpt-image-2.5-ext`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint. README excerpt: # GPT Image 2.5 API (`gpt-image-2.5-ext`) <!-- conv-kit:v1 --> <p align="center"> <img src="assets/badges/price.svg" alt="observed unit price"> <img src="assets/badges/billing.svg" alt="billing model"> <img src="assets/badges/compat.svg" alt="OpenAI-compatible endpoint"> </p> <p align="center"> <img src="assets/01-product-hero-perfume.jpg" width="820" alt="Sample render via the AP

### Overview: Server Nexe is a local AI server that runs entirely on your machine, providing persistent memory, RAG (Retrieval-Augmented Generation), and multi-backend inference capabilities. It is designed for full control and zero data sent to external services, making it ideal for developers and researchers. ### Key features: - Persistent memory for storing large datasets. - RAG for generating contextually relevant responses. - Support for multiple backend inference engines: MLX, LLaMA.cpp, and Ollama. - Runs on your machine, eliminating the need for cloud services. ### Tech stack: - Python ### Why it matters: Server Nexe offers a powerful and flexible solution for developers who need AI capabilities without the need for cloud services. It provides full control over the AI infrastructure, ensuring data privacy and security.

Overview: This project converts flowchart images into Mermaid code using a vision-language model called Llama Vision 11B. It supports both fine-tuned (FT) and non-fine-tuned (NFT) model evaluation. Key features: - Fine-tuned (FT) model evaluation - Non-fine-tuned (NFT) model evaluation - Converts flowchart images into Mermaid code - Supports both GPU and CPU - Uses the Unsloth and PEFT libraries Tech stack: Python Why it matters: This project provides a practical tool for converting flowchart images into code, which can be useful for various applications such as software development, process documentation, and more. Repository: jorgemunozl/Finetunning-Llama-Vision-11b GitHub stars: 11 GitHub description: Inference and finetunning of a VLM (LLama Vision 11b) using the Unsloth, PEFT libraries and torch. README excerpt: # Flowchart-to-Mermaid Transformer Thanks to Joel Timana for his support during the development of this project. This project converts flowchart images into [Mermaid](https://mermaid-js.github.io/) code using a vision-language model [Llama 3.2 Vision](https://huggingface.co/unsloth/Llama-3.2-11B-Vision-Instruct). It supports both fine-tuned (FT - LoRA adapters) and non-fine-tuned (NFT) model evaluation. Thanks to Unsloth

Overview: This project provides a Docker image for PostgreSQL that integrates PostGIS and pgvector, enabling efficient spatial data and vector similarity operations. Key features: - PostGIS extension for storing and performing spatial data operations - pgvector extension for high-dimensional vector similarity search Tech stack: Dockerfile Why it matters: This solution simplifies the integration of spatial data and vector similarity functionalities in PostgreSQL, making it easier for developers to work with spatial data and perform efficient similarity searches.

Overview: This project provides a GPT Image 2.5 Sunburst API endpoint for generating and editing images, offering a 20% discount for membership. Key features: - Generates and edits images with Sunburst format - Supports up to 16 references and PNG, JPEG, or WebP output - Offers a 20% discount for API access - Supports multiple image sizes and resolutions - Supports asynchronous processing Tech stack: Python Why it matters: This API is useful for developers looking to integrate image generation and editing capabilities into their projects, offering a cost-effective solution with a 20% discount for API access. Repository: joxsg/gpt-image-2.5-sunburst-api-fr Primary language: Python GitHub stars: 11

### Overview: FLUX 2 Pro API (`flux-2-pro`) is a model id, pricing, and parameterized API designed for OpenAI-compatible endpoints. It offers a single endpoint for generating images at $0.024 per image, with pricing based on resolution. The API supports various output sizes and resolutions, making it versatile for different needs. ### Key features: - **Model ID and Pricing**: FLUX 2 Pro generates images based on the provided model ID, with pricing per megapixel. - **Pricing Model**: Prices are based on the resolution of the output image, ranging from $0.024 for 1MP to $0.06 for 4MP. - **Model Page**: Access the detailed model page for specific pricing tiers and output sizes. - **API Key**: Obtain an API key to use the API, and the API is compatible with OpenAI endpoints. - **Pay-as-you-go**: No upfront costs, pay-as-you-go pricing model. - **Asynchronous Processing**: Tasks are processed asynchronously, with the cost reported in the response. ### Tech stack: Python ### Why it matters: - **Key Model**: The same Base-URL and `Authorization`-Header can be used for multiple models, reducing the need for switching clients. - **Cost Efficiency**: The API offers a single endpoint for generating images, with a cost of $0.024 per image, making it cost-effective. - **Dynamic Pricing**: Pricing is based on the resolution of the output image, ensuring that the cost scales with the