Projects

Overview: Vektori Memory is a personal AI memory extension that saves and searches conversations across multiple AI platforms, enhancing productivity and context. Key features: - Auto-save conversations from 6 major AI platforms - Semantic search across all saved conversations - Context injection to bring relevant memories into new chats - Privacy-first: your data is yours, authentication required Tech stack: JavaScript Why it matters: Vektori Memory streamlines communication by consolidating and searching across various AI platforms, making it easier and more efficient to recall information and stay productive. Repository: bandr-ai/vektori-extension Primary language: JavaScript GitHub stars: 11 GitHub description: Never repeat yourself across AI :)

Overview: This project fine-tunes Gemma-2 2B for structured data extraction using llama-cpp, aiming to improve the model's performance on extracting data into a structured format. Key features: - Fine-tuning and quantization of Gemma-2 2B using llama-cpp - Flask web server for serving the quantized model - OpenAPI schema for input/output data Tech stack: Jupyter Notebook Why it matters: This project provides a practical solution for extracting structured data from text, leveraging advanced fine-tuning techniques to enhance model performance. Repository: bastienpo/unsloth_finetuning Primary language: Jupyter Notebook GitHub stars: 11

### Overview The AI Companion is a versatile AI companion app that leverages generative AI models to provide chatbot services and AI-powered image, video, and audio generation. It integrates user personas and character personas to create genuine interactions, making AI more than just a tool. ### Key Features - **Chatbot**: Interact with AI using Large Language Models (LLMs). - **Image Generation**: Create images using models like Stable Diffusion and Flux. - **Audio Generation**: Generate audio content using AI models. - **Character Persona Settings**: Customize chatbot interactions with predefined prompts and presets. - **Hyperparameter Settings**: Adjust hyperparameters for better response quality. ### Tech Stack - Python ### Why It Matters The AI Companion transforms AI beyond mere tools into genuine friends and partners, making interactions more engaging and personalized. It caters to both general users and professional users, offering a wide range of applications from chatbot services to image and audio generation. ### Repository The repository is maintained by [bean980310](https://github.com/bean980310) and is hosted on GitHub. It has received 11 GitHub stars, indicating its popularity and usefulness in the developer community.

### Overview The Super AI Agent Companion is a versatile AI companion app that integrates generative AI models for chatbots, image generation, text generation, and audio creation. Ideal for both general users and professional professionals, it offers engaging persona-based interactions and creative content generation. ### Key Features - **Chatbot**: Utilizes Large Language Models (LLMs) for conversational AI. - **Image Generation**: Generates images using models like Stable Diffusion and Flux. - **Text Generation**: Produces text outputs from generative AI models. - **Audio Generation**: Creates audio content using AI-generated sounds. ### Tech Stack Python-based Gradio app leveraging generative AI models. ### Why It Matters The Super AI Agent Companion transforms AI into a true companion, enhancing productivity and creativity through personalized interactions and artistic content creation.

Overview: Vectaurant is an AI restaurant assistant that leverages OpenAI, Semantic Kernel, Qdrant, and the RAG pattern to provide a more natural and efficient interaction with menus and services. Key features: - 🌐 Natural Language Interaction: Customers can ask questions like "What’s spicy on the menu?" for smart, helpful answers. - 🧠 AI-Powered Agent with Plugins: Uses OpenAI models and Semantic Kernel to understand intent and execute tasks through custom plugins. - 📦 Vector Search with Qdrant: Menu items and restaurant knowledge are embedded using OpenAI embedding models and stored in Qdrant for fast semantic search. Tech stack: .NET Core, OpenAI API, Qdrant, Semantic Kernel. Why it matters: This project demonstrates the effective use of AI and semantic search to enhance user experience in a restaurant setting, making interactions more natural and efficient.

Overview: Tiny LLM is a small language model built from scratch in PyTorch, loaded with weights from GPT-2, Llama, Qwen2.5, and Pythia, and fine-tuned with LoRA for chat functionality. Key features: - Loads pre-trained weights from GPT-2, Llama, Qwen2.5, and Pythia - Fine-tunes with LoRA for enhanced performance - Implements a chatbot interface for interaction with the model Tech stack: PyTorch Why it matters: This project serves as an educational tool to understand and build large language models from scratch, focusing on the learning process rather than production-grade solutions. Repository: benisalla/tiny-llm GitHub stars: 10 GitHub description: A small LLM built from scratch in PyTorch. Loads GPT-2, Llama (SmolLM2), Qwen2.5 and Pythia weights with outputs identical to Hugging Face. LoRA fine-tuning, KV cache, Kaggle/Colab notebook and a Streamlit chatbot.

Overview: Picchio is a streaming Mixture-of-Experts (MoE) inference engine written in pure C, designed to run models larger than your RAM on ordinary consumer hardware. It optimizes by keeping only the dense part of the model in memory and streaming the experts from disk on demand, making it possible to run models like GPT-OSS (20B and 120B), Qwen3-MoE, and MiniMax-M2 (230B) comfortably on a 16 GB laptop. Key features: - Pure C implementation - Runs models larger than RAM - Streams experts from disk on demand - Optimized for streaming CPU Tech stack: Pure C Why it matters: Picchio efficiently handles models larger than RAM, making it suitable for running large models on consumer hardware without the need for expensive datacenter GPUs. This makes it ideal for applications requiring high throughput and low latency, such as real-time inference or streaming services. Repository: benmaster82/picchio GitHub stars: 11 GitHub description: A streaming Mixture-of-Experts (MoE) inference engine written in pure C, that runs models larger than your RAM on ordinary consumer hardware - GPT-OSS (20B and 120B), Qwen3-MoE, and MiniMax-M2 (230B).

Overview: This repository contains supplementary materials for the article "Creating Stable Diffusion 2.0 Service With BentoML And Diffusers," focusing on examples of deploying models using BentoML and Diffusers. Key features: - `sd2/` - Service with `txt2img/` endpoint using `stabilityai/stable-diffusion-2` - `sd2_mega/` - Service with `txt2img/` and `img2img/` endpoints using `stabilityai/stable-diffusion-2` and a custom Diffusers pipeline - `anything_v3/` - Service with `txt2img/` endpoint using `Linaqruf/anything-v3.0` Tech stack: Python Why it matters: This repository provides practical examples for deploying stable diffusion models using BentoML and Diffusers, making it valuable for developers looking to integrate these tools into their projects.

### Overview: Grok Imagine Image 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-hero-perfume.jpg" width="820" alt="Sample render via the APIMart image API"> </p> > **from $0.016 per image** 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-7482ca)** · **[Live pricing](https://go.apimart.ai/k-ad5527)** · **[Model page](https://go.apimart.ai/k-ef03ee)** · [⚡ 60-second quickstart](#quickstart) **Por que os times usam Grok Imagine Image (`grok-imagine-image`) via APIMart** - **Uma chave, o catálogo inteiro.** A mesma base URL `https://api.apimart.ai/v1` e o

Overview: This project involves finetuning Google's Gemma Model to enable natural language queries for translating into SQL, making it easier to interact with SQL databases using a user-friendly interface. Key features: - Natural language queries to SQL databases - Utilizing Google's Gemma Model for efficient translation - Step-by-step tutorial for setting up a powerful SQL querying system Tech stack: Jupyter Notebook Why it matters: This project simplifies the process of querying SQL databases by translating natural language queries into SQL, making it accessible and intuitive for users. Repository: bhattbhavesh91/google-gemma-finetuning-n2sql GitHub stars: 11 GitHub description: Finetuning Google's Gemma Model for Translating Natural Language into SQL README excerpt: <h1>Natural Language to SQL using Google's Gemma</h1> <p>Make natural language queries to a SQL Database using Google's Gemma Model. In this video, you will discover how you can harness the power of Google's Gemma Model to query a SQL Database using natural language. This step-by-step tutorial will guide you through setting up a powerful, intuitive, and user-friendly SQL querying system.</p> <h2>Click to open the Notebook directly in Google Colab</h2> <!-- [](https://colab.research.google.com/github/bhattbhavesh91/google-gemma-finetuning-n2sql/blob/main/n2sql-google-g

### Overview: Qwen-Image 3.0 is a powerful API that allows developers to generate and edit images using Alibaba Cloud's Qwen image route. It offers text-to-image capabilities with resolutions of 1K and 2K, and supports various parameters for customization. ### Key Features: - **Text-to-Image Generation:** Generates images from text prompts. - **Image Editing:** Allows editing of existing images. - **Multiple Outputs:** Supports generating multiple images per task. - **Flexible Resolution:** Outputs at 1K or 2K resolution. - **Model ID and Pricing:** Uses the `qwen-image-3.0` model and charges based on the number of images generated or the resolution tier. ### Tech Stack: - Python ### Why It Matters: - **Cost Efficiency:** Charges per image or resolution tier, making it cost-effective for large-scale projects. - **Model Compatibility:** Works seamlessly with OpenAI-compatible endpoints, ensuring compatibility with existing projects. - **API Key:** Access to the Qwen-Image 3.0 API via an API key, allowing for secure and efficient use. - **Quickstart:** Provides a simple command-line interface for developers to get started quickly. ### GitHub Repository: - **Repository:** biqwozw9wumgqu/qwen-image-3-0-api-ko - **Stars:** 11 - **Description:** Qwen-Image 3.0 API (`qwen-image-3.0`): model id, pricing, parameters, and copy-paste examples for the OpenAI

### Overview: This repository contains tested image prompt recipes for the FLUX Kontext model, including 8 recipes with example renders, aspect ratios, and the cost per image. The recipes are designed to help users create high-quality images efficiently. ### Key features: - **8 Recipes**: Eight tested image prompt recipes with example renders, aspect ratios, and the cost per image. - **Model ID**: `flux-kontext-pro` - **Cost per Image**: $0.032, estimated cost for the entire set: $0.256 ### Tech stack: Python ### Why it matters: These recipes are designed to help users create high-quality images efficiently, ensuring that the output meets the desired aesthetic and context. The cost per image is $0.032, making it an economical solution for users looking to optimize their image creation process.

### Overview: This repository contains tested image prompt recipes for the Qwen-Image 2.0 model, including examples, aspect ratios, and the associated costs. It provides a comprehensive guide for developers looking to optimize their image prompts. ### Key Features: - **Model ID:** `qwen-image-2.0` - **Aspect Ratios:** 1:1, 4:3, 3:2, 8:5, etc. - **Costs:** $0.02 per image ### Tech Stack: Python ### Why It Matters: - **Optimized Image Prompt Recipes:** Helps developers create high-quality images with minimal effort. - **Cost Efficiency:** The cost per image is $0.02, making it a cost-effective solution for developers. - **Versatile Applications:** Suitable for various use cases, including product photography, food photography, and more. ### Repository: Repository: `bx50l9qa/qwen-image-2-0-prompts-fr` GitHub Stars: 11 GitHub Description: "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." ### Example Recipes: - **Product / E-commerce:** - Model: `qwen-image-2.0` - Ratio: 1:1 - Cost: $0.02 - Prompt: "Studio hero shot of a frosted glass perfume bottle on a wet black

Overview: Cradle2Cane is a two-pass framework for high-fidelity face aging, addressing the trade-off between age accuracy and identity preservation. It uses adaptive noise injection and identity-aware embeddings to achieve better results across various age gaps and poses. Key features: - Adaptive noise injection mechanism (AdaNI) for controlling the strength of aging. - Two identity-aware embeddings (SVR-ArcFace and Rotate-CLIP) for enhancing identity preservation. - Joint training of both passes to improve overall performance. Tech stack: Python Why it matters: Cradle2Cane offers a more balanced approach to face aging, making it suitable for a wide range of applications, including entertainment and healthcare, by effectively managing the trade-off between age accuracy and identity preservation.

### Overview: Seedance 1.0 Pro Quality prompts offer tested image prompt recipes, including 4 recipes with example renders, aspect ratios, and model IDs. The cost per image is $0.0184 per second, making it an affordable and efficient way to generate high-quality images. ### Key Features: - **4 Recipes**: 4 tested recipes with example renders, aspect ratios, and model IDs. - **$0.0184 per Second**: Effective price used, estimated cost of the whole set: $0.0736. - **Model ID**: `doubao-seedance-1-0-pro-quality`. - **Duration and Cost**: Duration and cost are provided for each recipe. - **Prompt Structure**: A cheat-sheet for structuring prompts, including subject, light, framing, setting, and finish. ### Tech Stack: Python ### Why It Matters: Seedance 1.0 Pro Quality prompts are an efficient and affordable way to generate high-quality images, making it a valuable tool for developers and designers looking to streamline their workflow.

### Overview: Grok Imagine 1.5 Edit API (`grok-imagine-1.5-edit-apimart`) This API provides reference-guided image editing services, allowing users to modify images at a cost per image. ### Key Features: - **Model ID and Pricing**: The API is priced per image, with a base price of $0.015 per image. - **Model Catalog**: Access to over 300 image, video, and language models via the OpenAI-compatible `https://api.apimart.ai/v1` endpoint. - **Pricing Model**: Prices are per image generated, with a default list price of $0.01875 but effective at $0.015. - **Billing**: The API charges for each completed task, with failed tasks being refunded. - **Asynchronous**: Tasks are processed in batches and retries are handled through background jobs. ### Tech Stack: - **Python** - The primary language used to develop and interface with the API. ### Why It Matters: - **Cost-Effective**: Ideal for users looking to edit images without the need for a subscription. - **Model Diversity**: Access to a wide range of image, video, and language models. - **Pricing Transparency**: Clear pricing model with real-time updates on the pricing page. ### Repository: c-h93bzyc38vo3/grok-imagine-1-5-edit-api-es ### GitHub Stars: 11 ### GitHub Description: Grok Imagine 1.5 Edit API (`gro

### Overview: FLUX.2 Flex prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`flux-2-flex`) and the cost per image. ### Key features: - **Model ID**: `flux-2-flex` - **Aspect Ratios**: 1:1, 4:3, 3:2, 85mm, 24mm, macro, overhead flat lay - **Cost per Image**: $0.04 ### Tech stack: Python ### Why it matters: Eight FLUX.2 Flex prompts, each with an example render, aspect ratio, and cost on the APIMart route. The cost per image is $0.04, and the entire set's estimated cost is $0.32.

Overview: Imagen 4.0 API is a powerful image generation service provided by APIMart, offering a single endpoint for Google's Imagen model at $0.04 per image. It supports various parameters and pricing models, making it ideal for developers looking to integrate image generation into their projects without the need for an ongoing subscription. Key features: - A single endpoint for Google's Imagen model - Pricing per image - Supports multiple parameters and resolutions - Compatible with OpenAI's API - Asynchronous processing Tech stack: Python Why it matters: This API streamlines the process of integrating image generation into applications, offering a cost-effective solution for developers without the need for a monthly subscription.

Overview: This repository contains 8 tested image prompt recipes for the Grok Imagine Image Quality model, including examples of renders, aspect ratios, and the cost per image on the APIMart route. Key features: - 8 Grok Imagine Image Quality prompts - Example renders, aspect ratios, and cost per image - Model: `grok-imagine-image-quality` - Effective price: $0.04 per image Tech stack: Python Why it matters: This repository provides a concise overview of effective image prompt recipes for the Grok Imagine Image Quality model, helping developers and designers create high-quality renders without wasting resources.

Overview: Decis is a self-hosted inference server for open decision models, implementing the TypeSafe System One API. It provides a single API endpoint and supports multiple interchangeable engines, with one Docker image per engine. Key features: - Runs out of the box with three commands - A Jev-like API - Supports multiple engines with engine-specific tags - Provides a playground with three games Tech stack: Python Why it matters: Decis simplifies the migration of decision models to a self-hosted inference server, making it easier to integrate with existing systems and reducing the need for model downloads and mounting volumes. Repository: chaitin/Decis Primary language: Python GitHub stars: 11 GitHub description: Self-hosted, Jev-compatible decision-model API — one /v1/systemone endpoint, open weights (Laya, kev), one Docker image per engine.