Projects

### Overview: VectorDB is an in-memory vector database that offers a variety of indexing algorithms, flexible similarity metrics, metadata filtering, and a RESTful API for easy integration. It is designed to be production-ready with Docker support, scalability, and extensibility. ### Key Features: - **Multiple Indexing Algorithms**: Linear, KD-Tree, and LSH (Locality Sensitive Hashing) - **Flexible Similarity Metrics**: Cosine similarity and Euclidean distance - **Metadata Filtering**: Custom metadata filtering - **Persistence**: Snapshot and restore functionality - **Thread-Safe**: Custom reader-writer locks for concurrent operations - **In-Memory**: Fast access with in-memory storage - **RESTful API**: Full CRUD operations via FastAPI - **Python SDK**: Native client library for seamless integration - **Embeddings API**: Integrated Cohere support for text embeddings ### Tech Stack: Python ### Why it Matters: VectorDB provides a robust and scalable solution for in-memory vector databases, making it ideal for applications requiring fast access to vector data and efficient handling of metadata filtering. Its modular design and extensible nature make it easy to integrate with various use cases and add new indexing algorithms as needed.

Overview: Seedance 2.0 Mini API is a cost-effective solution for generating videos, providing a base URL and `Authorization` header to access the OpenAI-compatible APIMart endpoint. Key features: - Supports multiple resolutions and pricing tiers - No need for pre-paid subscriptions - Real-time pricing and cost reporting - Model id and route details - Model page for detailed information - Request parameters including model, prompt, duration, resolution, and seed Tech stack: Python Why it matters: Seedance 2.0 Mini API offers a cost-effective alternative to traditional video generation methods, providing a simple and efficient way to generate videos without the need for pre-paid subscriptions.

Overview: FLUX 2 Pro API, also known as `flux-2-pro`, is a model ID, pricing, and parameters-based OpenAI-compatible API endpoint for image generation. It offers a single endpoint compatible with OpenAI, allowing for easy integration and cost management. Key features: - Supports multiple image, video, and language models - Pricing per megapixel, with tiered options - Assynchronous execution, returning task IDs for tracking - Real-time pricing updates via the API Tech stack: Python Why it matters: This API streamlines the process of generating images from text or references, offering a cost-effective solution for developers looking to integrate OpenAI-compatible image generation into their projects without the need for a monthly subscription. GitHub stars: 12

Overview: Nano Banana 2 Lite API is a cheaper Gemini 3.1 Flash Lite Image route priced per image, offering a single key for all catalog access and a compatible OpenAI endpoint at `https://api.apimart.ai/v1`. Key features: - Single key for all API access - Per-image pricing - Compatible with OpenAI endpoint - Supports up to 300+ models and 300+ languages - Asynchronous processing from the start - Billing is reported in the response Tech stack: Python Why it matters: Nano Banana 2 Lite is a cost-effective alternative to Gemini 3.1 Flash Lite Image, providing a single key for all API access and a compatible OpenAI endpoint, making it ideal for developers looking to integrate with the APIMart endpoint without the need for a monthly subscription.

Overview: Grok Imagine Image Quality API is a powerful tool for generating and editing images using text prompts and natural language instructions. It offers a single OpenAI-compatible endpoint for unsubscription-based pricing, making it ideal for developers and businesses looking to integrate image generation into their applications. Key features: - Generates stunning photorealistic images from text prompts - Edits existing images with natural language commands - Supports multiple image resolutions and sizes - Charges per image delivered, with tiered pricing based on resolution - Asynchronous processing with task IDs and polling for status Tech stack: Python Why it matters: Grok Imagine Image Quality API simplifies the process of integrating image generation into applications, offering a single endpoint for unsubscription-based pricing, making it cost-effective and easy to integrate. The API supports multiple image resolutions and sizes, allowing for flexibility in image output quality.

### Overview: This repository contains 8 tested image prompt recipes for the Grok Imagine Image Quality model, along with examples of renders, aspect ratios, the model ID, and the cost per image. The API key is required for access. ### Key features: - 8 image prompt recipes - Example renders, aspect ratios, model ID, and cost per image - Accessible through an API key ### Tech stack: Python ### Why it matters: These recipes are designed to help developers create high-quality images efficiently, ensuring that the output meets the desired aesthetic and practical needs.

Overview: This project is a simple and modern web interface for LLM models based on LLaMA, designed for easy and convenient use. Key features: - Simple and modern Gradio web interface - Support for GGML and GGUF format (.bin and .gguf) models - Compatible with OpenAI and MistralAI APIs - Preset support for various models and endpoints - Easy installation and setup Tech stack: Python Why it matters: This web interface provides a user-friendly and efficient way to interact with LLM models, making it accessible and easy to use for developers and enthusiasts alike. Repository: ehristoforu/TensorLM-webui Primary language: Python GitHub stars: 12 GitHub description: Simple and modern webui for LLM models based LLaMA.

Overview: Jarvis is a simple voice assistant app built using Python, designed to perform various tasks through natural language queries and integration with GPT capabilities. Key features: - Voice recognition for commands - Wake word detection: "Jarvis" - Processing natural language queries - Integrating with chatGPT for responses - Performing internet searches - Sending email reminders - Creating email drafts Tech stack: Python Why it matters: Jarvis offers a user-friendly interface for executing tasks through voice commands, making it accessible and convenient for users. Its integration with GPT allows for intelligent responses, enhancing the overall functionality of the app. Repository: eliaswestonfarber/Jarvis Primary language: Python GitHub stars: 12 GitHub description: Voice Assistant with GPT abilities

Overview: CalGist is an AI-powered Google Calendar events summarizer built with Next.js, Supabase, and Google Gemini. It connects with your Google Calendar, generates summaries, and displays everything in a clean dashboard. Key features: - User Authentication - Google Calendar Integration - Upcoming Events View - AI-Powered Summaries - Clean Dashboard - Regenerate Summary Button - Copy to Clipboard Tech stack: Next.js + TypeScript, Supabase (Auth + Database), Google Calendar API, Google Gemini API, Tailwind CSS + ShadCN, Vercel for deployment. Why it matters: CalGist simplifies managing your Google Calendar by providing a user-friendly dashboard and AI-generated summaries, making it easier to stay organized and informed.

Overview: Experiment is an experimental chat interface for Large Language Models (LLMs) with function calling visualization support. Key features: - Multi-Provider Support - Advanced Tool Debugging - Include Context - Prompt Engineering - Import CSV - Universal Platform - Mobile Optimized - Secure by Design - OLED-Friendly - Free as in MIT - Privacy First Tech stack: TypeScript Why it matters: Experiment is an innovative exploration of how next-generation isomorphic web app architecture could look like, using experimental, bleeding edge stack. Repository: experimental-setup/Experiment Primary language: TypeScript GitHub stars: 12 GitHub description: 🔬 Experiment is an experiment is an experiment is an experiment is an experiment is an e̴x̷p̵e̵r̵i̶m̸e̵n̵t̴ ̷i̵s̵ ̷a̶n̶ ̷è̶̜x̵̨̎p̵̨̨̪̞̺̲̘̖̲̪̲̬̲̘̖̲̪̲̬̲̘̖̲̪̲̬̲̘̖̲̪̲̬̲̘̖̲̪̲̬̲̘̖̲̪̲̬̲̘̖

Overview: This project is a fine-tuning framework for Suiren molecular models, designed to adapt pre-trained models for various downstream property prediction tasks. The Suiren-ConfAvg model, which learns conformational averaging features, is particularly useful for tasks with limited data. Key features: - Fine-tuning models for molecular property prediction - Supports regression and classification tasks - Uses PyTorch for training and evaluation - Includes a complete set of fine-tuning models and training framework - Predefined or random train/val split options for data organization Tech stack: Python Why it matters: This framework provides a practical solution for molecular property prediction tasks, especially when dealing with limited data, making it valuable for researchers and developers working in the field. GitHub stars: 12

### Overview: Core DeepInfer is a Python library designed to trace and inspect the state transitions of a deep learning model during inference. It allows developers to follow a request from the initial token to the final committed token, including prefix recovery, speculative writes, and rollback operations. The project is built using NumPy and aims to provide a readable and practical tool for studying the state transitions of DeepInfer models. ### Key Features: - **Prefill and Prefix Recovery**: Tracks the restoration of a cached prefix. - **Verification Budgets**: Allocates and verifies tokens for model outputs. - **Speculative KV Rollback**: Handles speculative writes and rollback operations. - **Real Float64 Operations**: Performs real float64 attention, projections, and feed-forward operations. - **Random and Untrained Parameters**: Parameters are random and untrained, focusing on state inspection rather than model optimization. ### Tech Stack: - **Python**: The primary programming language used. - **NumPy**: Used for numerical operations and state management. ### Why It Matters: Core DeepInfer provides a detailed and readable way to trace the state transitions of deep learning models during inference, making it useful for developers and researchers looking to understand and debug the state of their models.

Overview: This project is a multimodal RAG demo based on Qwen3-VL models, designed to efficiently index and retrieve information from PDFs, images, and text. It includes a web interface for easy ingestion and querying. Key features: - Multimodal embeddings with Qwen3-VL - Multimodal reranker support - Option to generate answers with source context via local or OpenAI-compatible API Tech stack: Python Why it matters: This project leverages Qwen3-VL models to provide a more comprehensive and accurate search experience for documents, images, and text, making it easier to find relevant information and contextually understand the context of the retrieved results. Repository: holisticon/multimodal-rag-demo GitHub stars: 12 GitHub description: 🧠🖼️📄 Multimodal RAG Demo based on Qwen3-VL Embedding and Reranker models

Overview: This project provides a Dockerized version of TensorFlow Serving, allowing developers to easily deploy and manage their models. Key features: - Dockerized TensorFlow Serving - Supports both Docker and Docker Compose - Built-in support for gRPC requests via `utils.py` Tech stack: Python Why it matters: This project simplifies the deployment and management of TensorFlow models, making it easier for developers to integrate machine learning into their applications. Repository: huyhoang17/tensorflow-serving-docker Primary language: Python GitHub stars: 12

### Overview: This repository provides a multimodal activity recognition system that captures and retrieves data from multiple sensors simultaneously. It supports various platforms and sensors, including Windows, macOS, iOS, and Android, as well as hardware-integrated sensors and external devices like Kinect and Fitbit. ### Key features: - Multi-sensor data collection - Activity recognition using multiple modalities - Support for Windows, macOS, iOS, and Android platforms - Compatible with hardware sensors and external devices ### Tech stack: C ### Why it matters: This project is essential for researchers and developers looking to create intelligent systems that can understand human activities through various sensors, enhancing user experience and data-driven decision-making in applications like smart homes and wearables. ### Repository: idearibosome/multimodal-activity-recorder ### GitHub stars: 12 ### GitHub description: Multimodal recorder for activity recognition ### Primary language: C ### Dependencies: Qt 5.9+, Kinect SDK v2.0, PySide2 ### Project files: MultiModalRecorderServer.pro, MultiModalRecorderClient.pro, MultiModalRecorderViewer.pro ### License: LGPLv3 with additional regulations ### Acknowledgement: Supported by IITP grant funded by the Korea government (MSIT) (R7124-16-0004, Development of Intelligent Interaction Technology Based on Context Awareness and Human Intention Understanding)

Overview: Kling 2.6 Motion Control API is a comprehensive solution for developers looking to integrate AI-driven motion control into their applications. This API offers a single endpoint compatible with OpenAI, enabling developers to access a wide range of image, video, and language models without the need for a monthly subscription. Key features: - A single endpoint for OpenAI-compatible API - Access to over 300 different models - Pricing based on usage, starting at $0.0571 per second - Asynchronous processing, with task IDs and polling for task status - Optional prompt, character orientation, and sound/watermark control - Supports both standard (720p) and professional (1080p) resolutions Tech stack: Python Why it matters: This API simplifies the integration of AI motion control into applications, providing a cost-effective solution for developers without the need for a monthly subscription. The single endpoint and flexible pricing model make it easy to scale and manage resources efficiently. Repository: il0xj8m/kling-v2-6-motion-control-api-fr Primary language: Python GitHub stars: 12 GitHub description: Kling 2.6 Motion Control API (`kling-v2-6-motion-control`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint. README excerpt: # Kling 2.6 Motion Control API (`kling-v2-6-motion-control`) <!-- conv-kit:v1 --> <p align="center"> <img src="assets

### Overview: This is a set of interactive tutorials designed to help Python developers quickly integrate InterSystems IRIS for Health into their applications. The tutorials cover loading CSV and FHIR data, querying it with Python and SQL, vector search, and building LangChain AI agents and MCP servers. ### Key Features: - Load CSV and FHIR data into InterSystems IRIS for Health. - Query data using Python and SQL. - Implement vector search for fast semantic matching. - Build LangChain AI agents and MCP servers. - Access InterSystems IRIS for Health with Python. ### Tech Stack: Jupyter Notebook ### Why it Matters: This tutorial series provides a quick and easy way to integrate InterSystems IRIS for Health into Python applications, making it accessible to developers without prior experience with ObjectScript. The tutorials cover essential features for querying, vector search, and AI agent development, making it a valuable resource for anyone looking to build applications with InterSystems IRIS for Health.

### Overview: RAG Chat Application is a framework designed for the systematic evaluation of retrieval strategies and prompt engineering in RAG systems. It features an interactive chat interface for document analysis, including vector search, semantic chunking, and summarization capabilities. The application is built using Python and React with Flask, leveraging LlamaIndex, Chromadb, and Ollama for the backend and React for the frontend. It includes comprehensive RAG evaluation systems with automated QA pair generation, human curation interface, and multi-metric evaluation. ### Key features: - 🚀 Interactive chat interface - 📁 Document upload and processing - 🔍 Retrieval-augmented responses - Vector search - Semantic chunking - Summarization - Tools - Document tool use - Data science tool use - 🗂️ Session management - 🚧 Configurable model settings - 🎨 Modern, responsive UI - 📊 Comprehensive RAG evaluation system - Automated QA pair generation - Human curation interface - Multi-metric evaluation (programmatic, LLM-based, and hybrid) with option for human eval - Detailed performance analytics ### Tech stack: - Backend: Flask, LlamaIndex, Chromadb, Ollama, Sentence Transformers, spaCy - Frontend: React, Vite, TailwindCSS, React Router, Headless UI, Lucide Icons ### Why it matters: Running a full suite of ablation experiments with 2-3 different models costs

Overview: This project is a PyTorch implementation of the paper "MMoT: Mixture-of-Modality-Tokens Transformer for Composed Multimodal Conditional Image Synthesis". It aims to enable the synthesis of composed multimodal conditional images by utilizing a transformer-based approach. Key features: - Composed multimodal conditional image synthesis - Mixture-of-Modality-Tokens Transformer Tech stack: PyTorch Why it matters: This project offers a novel method for generating images that incorporate multiple modalities and conditional information, which could have applications in various fields such as computer vision and image synthesis.

### Overview: This project is a graduation project from Pusan National University, focusing on a food recommendation chatbot. The bot is trained using LLM fine tuning and utilizes a chatbot model named "ChatFoodie" (KoAlpaca Polyglot-5.8B-v1.0). The project includes a chat interface, a server for WebSocket API, and a front-end implementation using TypeScript and Next.js. The system is designed to be accessible with a non-member limit of 20 chats per day. ### Key Features: - **Chatbot Interface**: A chat interface for users to interact with the chatbot. - **WebSocket API**: A WebSocket server for real-time chat functionality. - **Server Implementation**: A Spring Boot server for handling requests and interacting with the chatbot. - **Frontend Development**: A Next.js front-end implementation for the chat interface. ### Tech Stack: - **Java**: Used for the backend server. - **Spring Boot**: Framework used for the server implementation. - **WebSocket**: For real-time chat functionality. - **TypeScript**: Front-end language used for Next.js. - **Next.js**: Framework for building the front-end. - **Tailwind CSS**: CSS framework for styling. - **Axios**: For making HTTP requests. - **MySQL**: Database management system for storing chat history. ### Why It Matters: This project showcases the development of a chatbot system that leverages LLM fine tuning for personalized food recommendations. The project's design and implementation highlight the integration of various technologies such as WebSocket, Spring Boot, and