Project directory

Projects

Open-source AI projects tracked by the leaderboard, ordered by GitHub stars.

Python 9 stars

pwmz6qjcw/gpt-image-1-5-api-ko

GPT Image 1.5 API (`gpt-image-1.5-official`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Overview: This repository provides detailed information on the GPT Image 1.5 API, including pricing, parameters, and usage examples. Key features: - Pricing details - Model id and routes - Request parameters - Quickstart guide Tech stack: Python Why it matters: This API allows developers to access OpenAI's GPT Image 1.5 model via a token-based billing system, offering flexibility and cost-effectiveness compared to traditional subscription models. GitHub stars: 9 Repository: pwmz6qjcw/gpt-image-1-5-api-ko

Unknown 9 stars

qdrant/qdrant-stars-handbook

The primary goal of Qdrant Stars is to recognize, reward, and support our most active and helpful users making significant contributions to the vector search community. The Qdrant Stars Handbook is your go-to resource for navigating the Stars program and how to take the most advantage of it.

### Overview: Qdrant Stars is a recognition program designed for the most engaged members of the Qdrant community. It provides exclusive resources, financial rewards, and support for content creation, and events. As a Qdrant Star, you gain access to beta program access, feedback loops, and product meetings, helping you shape the future of vector search technology. ### Key Features: - **Beta Program Access**: First access to Qdrant’s newest features and beta products. - **Feedback Loop**: Your feedback helps shape Qdrant’s development. - **Product Meetings**: Exclusive R&D sessions with the Core team. - **Content Creation**: Resources for creating informative and engaging content. - **Meetups and Workshops**: Opportunities to share knowledge and collaborate with peers. - **Engagement and Leadership**: Support for leading educational events and becoming a leader. ### Tech Stack: Qdrant Stars is built using the latest technologies and frameworks to ensure a seamless and efficient user experience. ### Why It Matters: Your contributions are essential in shaping the future of vector search technology, and your impact is recognized through exclusive resources, financial rewards, and direct access to Qdrant’s development team.

Python 9 stars

rguc5r0zdtlm/nano-banana-2-lite-prompts-ko

나노바나나 / 나노 바나나 — Nano Banana 2 Lite prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`gemini-3.1-flash-lite-image-ext`) and the cost per image.

### Overview: Nano Banana 2 Lite Prompts: 8 Tested Recipes with Outputs ### Key Features: - **Model ID:** `gemini-3.1-flash-lite-image-ext` - **Cost Per Image:** $0.0125 - **Aspect Ratios:** 1:1, 4:3, 16:9 - **Prompt Structure:** Structured in JSON format (`data/prompts.json`) and Markdown table (`PROMPTS.md`) ### Tech Stack: - Python ### Why It Matters: Nano Banana 2 Lite Prompts provide tested image prompt recipes with example renders, aspect ratios, the model ID, and the cost per image. This set includes eight recipes that follow three key rules to ensure a usable and efficient call to the Nano Banana 2 Lite model.

Python 9 stars

ritesh-modi/embedding-hallucinations

This repo shows how foundational model hallucinates and how we can fix such hallucinations using fine-tuning them

Overview: This repository focuses on understanding and mitigating hallucinations in embeddings, specifically how foundational models can fail to capture human-like understanding. It provides tools for fine-tuning SentenceTransformer models to reduce such hallucinations and improve semantic understanding. Key features: - Fine-tuning SentenceTransformer models to reduce hallucinations and improve semantic understanding. - Measuring similarity between sentences using cosine similarity and other metrics like dot product and Euclidean distance. Tech stack: Python Why it matters: This project is crucial for researchers and developers looking to enhance the interpretability and accuracy of embeddings in natural language processing tasks, ensuring that models better reflect human-like understanding and reduce the risk of hallucinations.

Python 9 stars

rm7p0q7p2/seedance-1-0-pro-prompts-de

Seedance 1.0 Pro prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`seedance-1.0-pro`) and the cost per image.

Overview: Seedance 1.0 Pro Prompts offers tested image prompt recipes, including 8 recipes with example renders, aspect ratios, the model ID (`seedance-1.0-pro`), and the cost per image on the APIMart route. Key features: - 8 tested image prompt recipes - Example renders and durations - Model ID: `seedance-1.0-pro` - Cost per image: $0.0184 per second Tech stack: Python Why it matters: Seedance 1.0 Pro (`seedance-1.0-pro`) provides a cost-effective and OpenAI-compatible endpoint for generating images, making it ideal for developers looking to integrate image generation into their projects without the need for a monthly subscription or prepaid plans.

Python 9 stars

rustyneuron01/Conversation-Genome-Project

Structured data & semantic tagging pipeline. Turns raw text (conversations, web pages, surveys) into tagged data for AI and search. Coordinators set ground truth; workers run LLM inference on windows. Scoring via cosine similarity. Python, FastAPI, OpenAI/Anthropic/OpenRouter, embeddings, Docker.

Overview: This project is a structured data and semantic tagging pipeline that converts raw text into tagged data suitable for AI and search applications. It uses low-cost, scalable semantic annotation and metadata generation with Large Language Models (LLMs) and embedding-based quality scoring. Key features: - Coordinators establish ground-truth tags and embeddings on full content, split it into windows, and score worker outputs. - Workers run LLM inference on text windows and return tags and embeddings. - Scoring uses cosine similarity between each worker’s output and the coordinator’s ground-truth tag embeddings (weighted combination of top-3 mean, mean, median, max; penalties for low overlap or too few tags). Tech stack: Python, FastAPI, OpenAI/Anthropic/OpenRouter, embeddings, Docker. Why it matters: This pipeline enables low-cost, scalable semantic annotation and metadata generation using LLMs and embedding-based quality scoring, making it suitable for various AI and search applications.

Python 9 stars

rx5umw-ict13/nano-banana-2-1-prompts-es

Gemini Nano Banana 2.1 prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`gemini-nano-banana-2.1`) and the cost per image.

Overview: This repository contains tested image prompt recipes for the Gemini Nano Banana 2.1 model, including 8 recipes with example renders, aspect ratios, and the cost per image on the APIMart route. Key features: - 8 tested image prompt recipes - Example renders, aspect ratios, and cost per image - Compatible with OpenAI's API endpoint Tech stack: Python Why it matters: This repository provides a practical guide for developers looking to optimize image prompts for the Gemini Nano Banana 2.1 model, ensuring accurate and cost-effective rendering.

C++ 9 stars

rye-001/Qwenium

Minimal C++ LLM Qwen/Gemma inference engine

Overview: Qwenium is a self-contained C++ engine designed for Qwen and Gemma families, providing fast, readable, and auditable inference capabilities. It supports up to 10 concurrent users and leverages a cross-family architecture, with attention materialized in the graph for easy access and efficient forward pass. Key features: - Multi-family support for Qwen 2/2.5/3/3.5/3.6/3.8 - Fast performance with metal throughout, batched multi-slot decode, and three tiers of KV reuse - Readable and changeable architecture with one directory per layer type and one file per model recipe - Auditable with weights hash, version-gated snapshots, kernel-path tags, and fail-loud replay refusal Tech stack: C++ Why it matters: Qwenium shows its work by exposing attention, determinism, and integrity as engine primitives, making it a portable and changeable solution for developers looking to integrate LLM inference engines into their projects.

Jupyter Notebook 9 stars

saimj7/Action-Recognition-in-Real-Time

A simple and fun video classification/action recognition using VGG16 as a feature extractor and RNN.

### Overview: This project utilizes VGG16 as a feature extractor and Recurrent Neural Networks (RNNs) for real-time video classification. The dataset consists of Cricket bowling and batting classes from UCF101, and the model achieves high accuracy in classifying video frames. ### Key Features: - **Feature Extraction**: VGG16 is fine-tuned on the dataset, and the top 4 layers are used as a feature extractor. - **Classification**: A simple classifier connects to the feature extractor, followed by a dense layer for sparse representations. - **Temporal Analysis**: Each video frame is represented as a tensor of size (NUM_FRAMES, LOOK_BACK, 1024), and the LSTM network is trained to distinguish between classes 1 and 2. - **Training**: The model is trained using a standard LSTM architecture, with GPU support required for CUDnnLSTM layers. - **Inference**: The model can be run on test videos or webcam, with inference times displayed and predictions sent via email if desired. ### Tech Stack: Jupyter Notebook ### Why it Matters: This project provides a simple yet effective solution for real-time video classification, which can be applied to various use cases such as anomaly detection and human action monitoring.

Python 9 stars

sethdford/gemma-realtime

Personalize Gemma 4 and make it real-time on Apple Silicon. Fine-tune on your conversations, serve at 111 tok/s. No cloud required.

Overview: Gemma 4 can be fine-tuned and served locally on Apple Silicon, providing real-time voice speeds without the need for cloud or API keys. Key features: - Fine-tune Gemma 4 with personalized conversations - Serve at 111 tokens per second (TPS) - Real-time voice conversation on Apple Silicon Tech stack: Python, MLX (optimized for inference), Gemma 4 E4B 4-bit quantization Why it matters: This project allows users to personalize their own conversations with Gemma 4, achieving real-time voice speeds without the need for cloud services, making it ideal for local use cases and accessibility.

Python 9 stars

sg40e98dutpek/wan-2-7-videoedit-prompts-fr

Wan 2.7 Video Edit prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`wan2.7-videoedit`) and the cost per image.

### Overview: This repository contains tested image prompt recipes for Wan 2.7 Video Edit, including 8 recipes with example renders, aspect ratios, and the model ID (`wan2.7-videoedit`). The cost per image is provided, along with a live pricing option and a model page. ### Key Features: - 8 tested recipes - Example renders - Aspect ratios - Model ID (`wan2.7-videoedit`) - Cost per image ### Tech Stack: Python ### Why It Matters: This repository provides practical tips and examples for developers looking to optimize their video editing workflows using Wan 2.7 Video Edit prompts.

JavaScript 9 stars

shekharP1536/ollama-web

Ollama Web UI is a simple yet powerful web-based interface for interacting with large language models. It offers chat history, voice commands, voice output, model download and management, conversation saving, terminal access, multi-model chat, and more—all in one streamlined platform.

Overview: Ollama Web is a web-based interface for interacting with large language models, offering features like chat history, voice commands, model management, and more. Key features: - Chat history - Voice commands - Model download and management - Conversation saving - Terminal access - Multi-model chat Tech stack: JavaScript Why it matters: Ollama Web simplifies the interaction with large language models by providing a user-friendly interface, making it accessible and efficient for developers and users alike. Repository: shekharP1536/ollama-web Primary language: JavaScript GitHub stars: 9 GitHub description: Ollama Web UI is a simple yet powerful web-based interface for interacting with large language models. It offers chat history, voice commands, voice output, model download and management, conversation saving, terminal access, multi-model chat, and more—all in one streamlined platform.

CSS 9 stars

shettysach/Protagonist-Producer

Protagonist Producer is an AI character creator app, powered by the OpenAI API. By typing in a brief idea, you get back a character portrait, a character name and a synopsis/backstory for your character using OpenAI API for Text-Davinci-003 and DALL·E 2.

Overview: Protagonist Producer is an AI character creator app powered by the OpenAI API, generating character portraits, names, and backstories based on your brief idea. Key features: - AI-generated character profiles - Character names - Backstory and synopsis Tech stack: Uses CSS for styling. Why it matters: Protagonist Producer streamlines the character creation process, making it accessible and efficient for anyone looking to create fictional characters.

Python 9 stars

shreyash321/Story-to-comic-AI

create any comic page using state-of-the-art text to image and large language models with your limitless imagination

Overview: Use your imagination to create unique comic pages with Story-to-comic-AI, a cutting-edge tool that leverages state-of-the-art text-to-image and large language models. Key features: - Generate unique comic pages from a single prompt - Harness the power of the state-of-the-art Palm language model API and Stable Diffusion XL model - Tailor every page to your specific prompt - No limitations on what types of comic pages you can create Tech stack: Python Why it matters: This AI-powered system allows users to unleash their creativity and generate custom comic pages without the constraints of traditional methods, making it a versatile tool for anyone looking to explore their imagination. Repository: shreyash321/Story-to-comic-AI Primary language: Python GitHub stars: 9 GitHub description: Create any comic page using state-of-the-art text to image and large language models with your limitless imagination. Our cutting-edge technology harnesses the power of the state-of-the-art Palm language model API and Stable Diffusion XL model to provide you with the ability to generate unique and custom comic pages from a single prompt. This means you can now unleash your imagination and create any type of comic page you desire, without the limitations of traditional methods. Our AI-powered system ensures that every page is unique and tailored to your specific prompt. You only need a Palm API and Hugging Face Stable Diffusion XL model API key to get started. Here's a simple comic page generated using Story-to-comic-AI: ![Untitled design (1)](https://github.com

Python 9 stars

simmia2026/SimMIA

The code for paper 'Membership Inference on LLMs in the Wild'

Overview: SimMIA is a robust Membership Inference (MI) framework designed for LLMs operating in a text-only regime, addressing the limitations of existing methods that rely on inaccessible model internals or poor generalization across domains. Key features: - Word-by-Word Sampling - Semantic Scoring - Relative Aggregation Tech stack: Python Why it matters: SimMIA achieves state-of-the-art results in the black-box setting, rivaling baselines that exploit internal model information, making it a crucial tool for auditing the training data of modern LLMs.

Jupyter Notebook 9 stars

singhdivyank/Data-Science-CheatSheets

Curated list of resources for Data Scientists, AI developers, and interview preperation

Overview: This repository is a curated list of resources for Data Scientists, AI developers, and interview preparation, featuring a curated collection of tools, techniques, and equations essential for success in the field. Key features: - A curated list of resources - Tools for Data Science and AI development - Equations and formulas for interview preparation Tech stack: Jupyter Notebook Why it matters: This repository serves as a valuable resource for anyone looking to improve their skills in Data Science and Artificial Intelligence, providing a comprehensive set of tools, techniques, and equations to help them succeed in their career.

Python 9 stars

smpanaro/token-recycling

Unofficial implementation of Token Recycling self-speculative decoding method.

### Overview: This is an unoffical implementation of the self-speculative LLM decoding method described in the paper "Turning Trash into Treasure: Accelerating Inference of Large Language Models with Token Recycling" by [Smpanaro](https://github.com/smpanaro/token-recycling). It aims to provide a fast and efficient way to decode large language models without requiring extra parameters or architecture changes. ### Key features: - Fast speedup over baseline [SpecBench](https://github.com/hemingkx/Spec-Bench) on A100. - Plug-and-play, no training or architecture changes needed. - Self-speculative decoding method. ### Tech stack: Python ### Why it matters: This implementation provides a fast and efficient way to decode large language models without requiring extra parameters or architecture changes, making it a valuable tool for developers looking to optimize their models without additional training or changes.

Python 9 stars

snsupratim/RagBot-2.0

### Overview: This project is a modular **Retrieval-Augmented Generation (RAG)** application that allows users to upload PDF documents and chat with an AI assistant that answers queries based on the document content. It features a microservice architecture with a decoupled **FastAPI backend** and **Streamlit frontend**, using **ChromaDB** as the vector store and **Groq's LLaMA3 model** as the LLM. ### Key features: - 📄 Upload and parse PDFs - 🧠 Embed document chunks with HuggingFace embeddings - 💂️ Store embeddings in ChromaDB - 💬 Query documents using LLaMA3 via Groq - 🌍 Microservice architecture (Streamlit client + FastAPI server) ### Tech stack: Python ### Why it matters: This project showcases the power of combining LLMs with external knowledge through RAG, providing more accurate and context-aware responses. The modular architecture allows for easy integration and deployment, making it suitable for various applications requiring document-based AI assistance.

TypeScript 9 stars

supaIC/Cipher-AI-Vault

Cipher AI Vault is an Azle-based proof of concept designed with a strong emphasis on abstraction, seamlessly integrating in-memory VectorDB, in-memory LLM, secure asset storage, stable memory data storage, cycles-distro top-up, and ic-auth for authentication.

Overview: Cipher AI Vault is a proof of concept Azle-based project that integrates various AI and blockchain technologies, including in-memory VectorDB, LLM, secure asset storage, stable memory data storage, cycles-distro top-up, and ic-auth for authentication. Key features: - In-memory VectorDB - In-memory LLM - Secure asset storage - Stable memory data storage - Cycles-distro top-up - ic-auth for authentication Tech stack: TypeScript Why it matters: This project showcases the potential of the Internet Computer for secure, sandboxed AI development, providing adaptable tools for a wide range of AI-driven applications. GitHub stars: 9 Repository: supaIC/Cipher-AI-Vault

Python 9 stars

svjack/ControlLoRA-Chinese

A Light Neural Network To Control Stable Diffusion Spatial Information tuned by Chinese

### Overview: ControlLoRA-Chinese 是一个使用中文微调的轻量级网络，旨在控制 Stable Diffusion 的空间信息。通过 LoRA 技术，该网络在不牺牲性能的情况下大幅减少了参数数量，从而提高了部署效率。该工程基于 GitHub 上的 ControlLoRA，提供两个中文领域的模型供用户使用。 ### Key features: - 使用中文微调的控制 Stable Diffusion 的空间信息。 - 采用轻量级网络 (~7M 参数个数, ~25M 存储)。 - 提供两个中文领域的模型供用户使用。 ### Tech stack: Python ### Why it matters: 通过 LoRA 技术，该网络在不牺牲性能的情况下大幅减少了参数数量，从而提高了部署效率。这对于需要频繁微调或切换任务的场景非常有用。