Projects

Overview: This project is an AI tool designed to analyze the layout grid systems of posters, allowing users to detect and view the grid system, export it as SVG guides or CSS Grid/Tailwind code, and even export it as SVG guides or CSS Grid/Tailwind code. The tool is model-agnostic, using a multimodal vision model to analyze grid systems, elements, keylines, and color palettes. Key features: - Vision-model analysis of grid systems, elements, keylines, and color palettes. - Web UI with live grid overlay, layer toggles, and an export dialog. - Export options for SVG grid guides (Figma/Illustrator ready) or CSS Grid/Tailwind code. - CLI (`npm run analyze -- ./poster.jpg --summary`) for no server required. - OpenAI-compatible multimodal model support. - Robust pipeline with timeout, retry, schema validation, normalization, and in-memory cache. - Agent-ready, ships as an installable skill (`SKILL.md`) and `AGENTS.md`. Tech stack: HTML. Why it matters: This tool simplifies the process of analyzing and understanding the layout of posters, making it easier for designers and developers to work with grid systems and optimize their designs. Repository: Ygria/poster-grid-inspector GitHub stars: 13 GitHub description: Model-agnostic AI poster grid inspector: detect a poster's layout grid system with a vision model, view it overlaid, and export SVG guides / CSS Grid / Tailwind code. Ships as an installable agent skill.

Overview: This project fine-tunes the T5 model on the SQuAD dataset to generate questions. Key features: - Fine-tuning T5 model - SQuAD dataset integration - Question generation capability Tech stack: Python Why it matters: This project demonstrates the application of T5 fine-tuning on a practical dataset, showcasing its potential for generating questions from text data.

Overview: AV-Phys Bench is a comprehensive benchmark designed to evaluate how well audio-video generation models understand and preserve physical commonsense as scenes evolve over time. It includes three categories: steady state, event transition, and environment transition, with an Anti-AV-Physics subcategory to test generative physics knowledge. Key features: - Comprehensive dataset for evaluating physical commonsense in joint audio-video generation - Seven generators evaluated by human raters, an MLLM-as-judge baseline, and the AV-Phys Agent - Live leaderboard, video gallery, and per-prompt rubric details on the project page Tech stack: Python Why it matters: This benchmark is crucial for advancing the field of audio-video generation and understanding how models interpret physical reality, providing valuable insights into the limitations and capabilities of current AI systems. Repository: ZijunCui02/AV-Phys Primary language: Python GitHub stars: 13 GitHub description: Project page for AV-Phys Bench: Do Joint Audio-Video Generation Models Understand Physics?

Overview: This project is a computer vision system designed to detect and recognize license plates in images or video streams. Key features: - Detects license plates from video streams. - Extracts alphanumeric characters from the detected license plates. - Associates detected plates with their respective vehicles. - Utilizes a custom-trained license plate detection model. - Reads license plate contents using the PaddleOCR English pretrained model. - Tracks vehicles using ByteTrack algorithm. Tech stack: Python Why it matters: This project is significant for applications that require accurate and efficient detection of license plate information, such as traffic monitoring, license plate recognition systems, or automated license plate recognition for security purposes. Repository: ahmedshafiq12/License-Plate-Recognition-System Primary language: Python GitHub stars: 13 GitHub description: The License Plate Recognition (LPR) System is a computer vision project designed to detect and recognize license plates in images or video streams.

Overview: LLM Sentinel showcases NVIDIA's NeMo-Guardrails and LangChain for enhancing the safety of Large Language Models (LLMs). Key features: - LLM Sentinel integrates NVIDIA's NeMo-Guardrails and LangChain for improved LLM safety. - It utilizes the Llama3:8b model for development and the Llama3:70b model for production environments. - The project is compatible with Python 3.9 and above, running on macOS and Ubuntu Linux. Tech stack: Python Why it matters: This project provides a practical demonstration of how to enhance the safety of LLMs using cutting-edge AI tools, making it valuable for developers and researchers in the field.

### Overview: ComfyUI Module is a one-click portable installer for ComfyUI, a high-performance AI art generation tool. It eliminates the need for Docker, Python, Git, or admin rights, automatically downloading and isolating everything into a single, relocatable folder. ### Key Features: - **Multi-GPU Support**: Automatically detects and runs ComfyUI on multiple GPUs simultaneously, allowing for parallel processing. - **Multi-Instance Management**: Easily manage multiple ComfyUI servers with one-click controls, each with its own GPU, port, VRAM mode, and log prefix. - **101 Pre-Built Models**: Covers all default ComfyUI workflows, including Flux 1.x, SDXL, SD 3.5, SD 1.5, LTX-2, HunyuanVideo, Wan 2.1, and more, with automatic downloads and sorting by category and VRAM tier. - **16 Curated Custom Nodes**: Essential and recommended extensions, installed with automatic dependency handling. - **SageAttention**: One-click install for significantly faster attention computation, especially beneficial for video generation. - **Zero Dependencies**: No Python, Git, CUDA toolkit, Docker, or admin rights required. Everything is downloaded and configured into a portable folder. ### Why This Exists: - Eliminates the need for multiple installations, configurations, and dependencies. - Automatically downloads, configures, and isolates everything, ensuring no system changes or admin rights required. - Provides a fully relocatable, zero-dependency installation. ### Tech Stack: - Python

### Overview: This project optimizes Qwen3.5 fine-tunes on NVIDIA DGX Spark, achieving a 7.67× LoRA speedup and an 8.35× Full Fine-Tuning speedup compared to renting a H100 GPU. The loss curve is verified to be lossless within BF16 precision, and the training job runs on real multimodal data with a wall-clock parity to a rented H100 GPU in just 20 hours. ### Key features: - **LoRA Speedup:** 7.67× faster than the stock Unsloth. - **Full Fine-Tuning Speedup:** 8.35× faster than renting a H100 GPU. - **Lossless Training:** The loss curve is verified to be lossless within BF16 precision. - **Real Data Training:** The training job runs on real multimodal data with a wall-clock parity to a rented H100 GPU. - **Interactive Wizard:** The wizard handles model picker, data validator, training, and merge in three commands. ### Tech stack: Python ### Why it matters: This project demonstrates the potential of using your existing hardware to achieve significant speed improvements in fine-tuning models, making it cost-effective and practical for developers and researchers.

Overview: This repository contains Docker Compose configurations for deploying Qwen3.8-27B on a DGX Spark (GB10) unit. The setup includes SGLang, NVFP4, and DFlash2, and is currently validated and optimized for high quality and speed. Key features: - SGLang + NVFP4 + DFlash2 - Production stack: Qwen3.8-27B + DFlash2 - Benchmarking validated on a clean clone - Re-validated on a clean clone - Rejected chat template: Froggeric v22.4 - BF16-LMHead rejected - Switched from vLLM+NVFP4+MTP to SGLang+NVFP4+DFlash2 Tech stack: Shell Why it matters: This setup ensures high quality and speed, making it suitable for applications requiring both robustness and performance.

### Overview: FeelingBlue is a corpus designed to explore the emotional connotation of color in context, presented at TACL 2022 and ACL 2023. It comprises 19,788 4-tuples of abstract art ranked by annotators according to their evoked emotions and paired with rationales for those annotations. ### Key Features: - **Corpus Size:** 19,788 4-tuples of abstract art. - **Annotations:** Each 4-tuple is ranked by an annotator according to their evoked emotions. - **Rationales:** For each selected image, a rationale is provided explaining why it is the least or most emotional image. - **Contextual Emotions:** The corpus focuses on the emotional connotation of color in context, exploring the intensity of perceived emotions. ### Tech Stack: - Python ### Why It Matters: FeelingBlue provides a rich dataset for researchers and practitioners interested in understanding how context affects the emotional connotation of color. By offering a baseline for a new task—Justified Affect Transformation—the corpus opens up new avenues for investigating the emotional impact of color in different contexts, potentially leading to advancements in color psychology and emotional design.

Overview: Aura Emotion AI is a sophisticated AI companion system that integrates advanced vector database, emotional intelligence, and model context protocol integration, providing a comprehensive and adaptive user experience. Key features: - Advanced cognitive architecture with ASEKE Framework - Tentative emotion assessment with checked source quotations and explicit uncertainty - Cognitive focus tracking across different mental frameworks - Adaptive self-reflection for continuous improvement - Thinking extraction with transparent AI reasoning - Intelligent memory system with Vector Database Integration, Persistent Conversation Memory, and MemVid AI QR code mp4 memory - MCP integration for Model Context Protocol Server and Client - Advanced analytics with Emotional Trend Analysis, Cognitive Pattern Recognition, and Personalized Recommendations - Data flow from user input to response generation, state updates, and memory storage Tech stack: Python Why it matters: Aura Emotion AI offers a robust and adaptable AI companion system that enhances emotional intelligence and memory management, making it a valuable tool for users seeking advanced cognitive support. Repository: angrysky56/emotion_ai GitHub stars: 13 GitHub description: The Aura Emotion AI system is a sophisticated AI companion that integrates advanced vector database, emotional intelligence, and model context protocol integration, providing a comprehensive and adaptive user experience.

### Overview: KnowledgeHub is an AI-enhanced development platform that integrates Claude Code, GitHub Copilot, and other AI tools, providing persistent memory, intelligent context, and advanced security features. It transforms your development workflow with 99.9% uptime and enterprise-level support for multi-project isolation and team collaboration. ### Key Features: - **Memory & Learning**: Remembers your development patterns and decisions, learns from errors and successful solutions, and maintains context across coding sessions. - **AI Enhancement**: Works with Claude Code, GitHub Copilot, Cursor, and any AI tool, providing intelligent context injection, suggesting relevant code patterns and solutions, and analyzing and improving development workflows. - **Enterprise Features**: Vector search across all your code and documentation, real-time analytics and performance monitoring, multi-project isolation and team collaboration, advanced RAG systems. ### Tech Stack: Python ### Why it Matters: KnowledgeHub transforms your development workflow with persistent memory, intelligent context, and advanced security & compliance, making it a valuable tool for enterprise-level developers. ### Repository: anubissbe/knowledgehub ### GitHub Stars: 13 ### GitHub Description: 🧠 Enterprise AI-Enhanced Development Platform with GraphRAG, LlamaIndex, Multi-Tenant Architecture, and Advanced Security & Compliance. Transform your development workflow with persistent memory, intelligent automation, and 99.9% uptime.

Overview: A developer-friendly package for setting up a Qdrant instance without Docker, ideal for development environments. Key features: - Runs Qdrant in a local environment - Supports multiple operating systems (Linux, macOS, Windows) - Automatically fetches the correct Qdrant binary for the user's platform and architecture - Sets up necessary configuration files and dashboard - Cleans up unnecessary node_modules after installation Tech stack: JavaScript Why it matters: This package simplifies the process of setting up a Qdrant instance, making it accessible for developers without the need for Docker, which can be a barrier in some development environments.

Overview: This project fine tunes a text generation model using the GPT-2 architecture and a CSV dataset. The code covers installation, dataset processing, tokenization, model import, training, saving, and inference steps. Key features: - Fine-tuning of a GPT-2 model - CSV dataset for training - Tokenization using GPT-2 tokenizer - Training using PyTorch and Hugging Face Transformers - Model saved to output directory - Inference capability with trained model Tech stack: Python, PyTorch, Hugging Face Transformers, GPT-2 tokenizer, Hugging Face Datasets Why it matters: This project provides a comprehensive guide for fine-tuning a GPT-2 model for text generation tasks, particularly useful for correcting HTML code, and can be adapted for various text generation applications.

### Overview: Mull-Tokens is a method designed to compress textual and visual reasoning information into modality-agnostic discrete latent tokens, enhancing multi-modal reasoning with Qwen2.5-VL. ### Key features: - Mull-Tokens Stage 1: Latent compression for modality agnostic reasoning. - Mull-Tokens Stage 2: Discrete latent learning for improved reasoning. - GRPO (Reinforcement Learning): A reinforcement learning approach to further enhance model performance. ### Tech stack: Python ### Why it matters: Mull-Tokens provides a unique way to integrate textual and visual reasoning information, making it easier to train Qwen-style models and evaluate them effectively using lmms-eval.

Overview: This project involves a semantic segmentation challenge using multimodal remote sensing earth observation data, focusing on land cover description. The dataset includes 70,000+ aerial imagery patches and 50,000 Sentinel-2 satellite acquisitions. Key features: - Real-world scenario exploration - Heterogeneous data fusion methods - 30 teams participated, with our team ranking 8th with a mIoU score of 0.62610 Tech stack: Jupyter Notebook Why it matters: This project provides a practical application of semantic segmentation techniques in remote sensing, showcasing the effectiveness of data fusion methods in real-world scenarios. Repository: association-rosia/flair-2 GitHub stars: 13

### Overview: Krea 2 Turbo is a 12.9B text-to-image model ported to Apple MLX, providing a faithful and faithful-to-PyTorch version of the model. This project offers a local web UI and command-line interface, ensuring a smooth and efficient experience on Apple Silicon Macs. ### Key features: - **Local Web UI**: A one-command web interface for generating images. - **Command Line Interface**: Supports command-line generation with various options for customization. - **8-bit and Mixed-4/8 Builds**: Available builds for different memory requirements and model quality. ### Tech stack: - **Apple MLX**: The underlying technology used for the project. ### Why it matters: This project provides a faithful and efficient version of Krea 2 Turbo, making it accessible and practical for developers and enthusiasts on Apple Silicon Macs. The local web UI and command-line interface ensure flexibility and ease of use, making it suitable for both beginners and advanced users. ### Repository: The repository is hosted on GitHub at [https://github.com/avlp12/krea2_alis_mlx](https://github.com/avlp12/krea2_alis_mlx). ### GitHub Stars: The repository has been starred by 13 developers, indicating its popularity and usefulness in the community. ### Primary Language: The project is written in Python, ensuring compatibility and ease of use for developers familiar with the language. ### GitHub Description: Krea 2 Turbo (12.9B text-to-image) ported to Apple MLX —

Overview: AvocadoDB is a deterministic context database designed for AI agents, ensuring the same query always returns the same context, improving accuracy and efficiency. Key features: - Deterministic query results - 6x faster compilation speed - Zero cost with pure Rust embeddings - Works offline and without internet - Citation-backed context generation - Token-efficient with 95%+ budget utilization - Drop-in replacement for existing RAG systems Tech stack: Rust Why it matters: AvocadoDB addresses the fundamental issues of current RAG systems, providing a stable, fast, and cost-effective solution that eliminates the variability and inefficiencies associated with traditional vector databases. Repository: avocadodb/avocadodb GitHub stars: 13

Overview: Wan 2.7 Image prompts offers tested recipes for image prompts, including aspect ratios, model IDs, and pricing information. Key features: - 8 tested recipes with example renders, aspect ratios, model IDs (`wan2.7-image`), and cost per image. - Compatible with OpenAI's API endpoint `https://api.apimart.ai/v1`. - Estimated cost of $0.1728 per image for the entire set. Tech stack: Python Why it matters: Wan 2.7 Image prompts provide a reliable and cost-effective way to generate high-quality image prompts, making it easier for developers to create visually appealing designs and marketing materials without the need for expensive subscriptions.

### Overview: This repository contains the implementation for the paper titled "Zero-1-to-3: Domain-level Zero-shot Cognitive Diagnosis via One Batch of Early-bird Students towards Three Diagnostic Objectives", published at AAAI 2024. ### Key Features: - Zero-shot Cognitive Diagnosis - Domain-level Diagnosis - One Batch of Early-bird Students - Three Diagnostic Objectives ### Tech Stack: Python ### Why it Matters: This project addresses the challenge of zero-shot cognitive diagnosis by leveraging early-bird students' data, making it particularly useful for educational applications where timely data is crucial.

### Overview: Nano Banana Pro Prompts: 8 Tested Recipes with Outputs ### Key features: - **Model ID:** `gemini-3-pro-image-preview` - **Cost per Image:** $0.03 (4K: $0.04) - **Aspect Ratios:** 1:1, 4:3, 8:5, 8:7, 16:9, 1:16, 2:3, 3:4 - **Effective Price:** $0.24 for the set ### Tech stack: Python ### Why it matters: Nano Banana Pro Prompts provide tested image prompt recipes with examples, aspect ratios, and the cost per image, making it easier for developers to create high-quality images without the need for expensive APIs or complex configurations.