Projects

Overview: Explore practical fine-tuning of Large Language Models (LLMs) using LoRa (Low Rank Adaptation). Key features: - Trainable A & B (only) - Efficient model adaptation across diverse tasks Tech stack: Python, PyTorch, Hugging Face Transformers Why it matters: This repository provides foundational examples for fine-tuning LLMs using LoRa, showcasing efficient model adaptation across various tasks.

Overview: This repository contains 9 n8n automation workflows showcasing advanced AI-powered automation, vector database integration, and multi-modal processing. It includes WhatsApp agents, RAG chatbots, sentiment analysis, email automation systems, and more. Key features: - WhatsApp Agent - Vector Database Integration - Sentiment Analysis - Respond to Sponsorship - RAG Chatbot - Google Sheets Integration - Form Submission to Airtable - Customer Service Assistant - AI-Powered Email Automation System Tech stack: n8n, OpenAI GPT-4, Pinecone, Google Sheets, Airtable, WhatsApp, Gmail, Google Drive, Pinecone, Vector Databases, n8n Features Why it matters: This repository demonstrates the power of AI and n8n in automating complex business processes and integrating with multiple platforms, making it a valuable resource for developers looking to enhance their automation workflows.

### Overview: This repository is a curated collection of the latest AI Engineer Development Tools, including databases, vector databases, data ingestion, data processing, data warehousing, model training, model zoo, model tuning, model evaluation, model deployment, model monitoring, and an AI agent framework. ### Key Features: - **Database**: MongoDB, MySQL, PostgreSQL, and Chroma, LanceDB, Milvus, Pinecone. - **Vector Database**: Chroma, LanceDB, Milvus, Pinecone. - **Data Ingest**: No mention. - **Data Processing**: No mention. - **Data Warehouse**: No mention. - **Model Training**: No mention. - **Model Zoo**: No mention. - **Model Tuning**: No mention. - **Model Evaluation**: No mention. - **Model Deployment**: No mention. - **Model Monitor**: No mention. - **AI Agent Framework**: No mention. ### Tech Stack: GitHub stars: 8 ### Why it Matters: This repository serves as a comprehensive resource for developers looking to enhance their AI engineering toolkit. It provides a centralized location for accessing the latest tools and technologies in the field, ensuring that developers can stay up-to-date with the most advanced AI development practices.

Overview: Context Engineering is the art of providing all the necessary context for a Large Language Model (LLM) to solve tasks effectively. It involves understanding and incorporating various elements like instructions, user prompts, long-term memory, retrieved information, available tools, and structured outputs into the conversation. Key features: - Instructions / System Prompt - User Prompt - State / History (short-term Memory) - Long-Term Memory - Retrieved Information (RAG) - Available Tools - Structured Output Tech stack: Python and various LLM integrations. Why it matters: Context Engineering ensures that the LLM has all the necessary information to understand and solve tasks efficiently, leading to more effective and accurate responses.

Overview: This repository contains the code and evaluations for the paper "Where a New Concept Must Enter: Entry Point Gates Cross-Task Usability in Unified Multimodal Models". It focuses on separating tasks in a unified multimodal model to understand how different pathways influence cross-task usability. Key features: - Novel entity and rendered Objaverse asset paired with a screened pseudo-word. - Separation of tasks by construction: understanding and generation are bound through one task direction, and the untrained direction is measured. - Closed-form activation edit (`src/train/hidden_ti.py`) makes a concept drawable at layer 7 of 28 with every weight frozen, and is indistinguishable from the base model from layer 14 on. - Weight-carried version of the same objective (`src/train/anchor.py`) peaks at layers 10–14 across four models. - The window appears only where the understanding pathway is a semantic vision encoder. Tech stack: Python Why it matters: This repository provides a detailed analysis of how different pathways in a unified multimodal model affect cross-task usability, offering insights into the design of such models and their potential for improving cross-task usability in multimodal tasks. Repository: Zane-ZYQiu/entry-point-umm Primary language: Python GitHub stars: 8 GitHub description: Code and evaluations for "Where a New Concept Must Enter: Entry Point Gates Cross-Task Usability in Unified Multimodal Models" ([arXiv:2608.17564](https://arxiv.org/abs/

Overview: Academic RAG Assistant is an AI-powered tutoring system that transforms your course textbooks into an interactive AI conversation partner, providing contextual answers across multiple subjects including Linear Algebra, Discrete Structures, and Calculus & Analytical Geometry. Key features: - Multi-Subject Expertise - Agentic RAG System - Real-time Streaming - Modern UI - Persistent Chat History - Support for Multiple Google Gemini Models - Error Handling - Progress Tracking - Smart Retrieval - Memory System Tech stack: Streamlit, Python, asyncio, OpenAI Agents SDK, Google Gemini API, Pinecone, SQLite, LangChain, Sentence Transformers, HuggingFace. Why it matters: This system enhances the learning experience by providing personalized, contextually relevant answers, making complex academic topics more accessible and engaging.

Overview: This repository contains the code used to generate the plots in a paper titled "Optimal Scheduling Algorithms for LLM Inference: Theory and Practice." It includes scripts to reproduce key figures from the paper and evaluate different scheduling policies. Key features: - Scripts to generate and plot median TTFT and batch execution time vs. token budget. - Scripts to generate GPU memory utilization vs. number of active requests. - Scripts to generate batch execution time vs. number of decode tokens. Tech stack: Python Why it matters: This repository provides a practical tool for evaluating and comparing scheduling algorithms for large language model inference, based on the insights from the paper's findings.

Overview: The Gemini Prompt Engineering Toolkit is a comprehensive application designed to streamline AI prompt engineering and fine-tuning dataset workflows. It utilizes Google's latest Gemini models to generate high-quality prompts, analyze files, and create synthetic datasets for AI model fine-tuning. Key features: - **Prompt Generation:** Craft effective prompts for various tasks, including creative writing and code generation. - **File Analysis:** Gain insights from uploaded and analyzed CSV, TXT, Markdown (MD), and image files. - **Test Data Generation:** Create synthetic datasets for fine-tuning AI models, ensuring optimal performance. - **Multi-Model Support:** Choose from Gemini-1.5-flash-exp-0827, Gemini-1.5-pro-exp-0827, and Gemini-1.5-flash-8b-exp-0827 models to leverage different capabilities. - **User-Friendly Interface:** Intuitive Streamlit interface for easy navigation and access. Tech stack: Streamlit, Python, and Gemini AI. Why it matters: The Gemini Prompt Engineering Toolkit is designed to help developers become proficient in AI prompt engineering, enabling them to fine-tune AI models effectively and streamline their workflows. By leveraging the latest Gemini models, it provides powerful tools for generating high-quality prompts, analyzing data, and creating synthetic datasets, making it a valuable resource for AI enthusiasts and professionals alike. Repository: aiengineeringforgrandmas/gemini-prompt-engineer-toolkit GitHub stars: 8 GitHub description: ⚡ Powered by the latest (Aug 27, 2024)

Overview: Aethelgard is a secure Federated Retrieval-Augmented Generation (FedRAG) framework designed to query highly sensitive distributed vector databases without moving raw data or opening inbound corporate firewalls. It eliminates the need for Data Use Agreements (DUA) and creates a scalable infrastructure for global clinical consensus. Key features: - Pure-Pull Architecture: Edge nodes use outbound asynchronous polling, eliminating the need for inbound port-forwarding. - 100% Pluggable: Ships with FastAPI and Redis defaults, but core abstractions allow easy swapping to gRPC, Kafka, AWS SQS, or GCP Pub/Sub. - Semantic Firewall Ready: Designed to easily integrate local LLM verification (e.g., MedGemma) to sanitize vector search results before transmission. Tech stack: Python Why it matters: Aethelgard addresses the rare disease "Diagnostic Odyssey" by providing a decentralized nervous system for clinical intelligence, enabling global clinical consensus without compromising patient privacy or data security.

Overview: MJ-Desktop - Midjourney AI image generator for Windows, featuring offline mode and batch processing, with a version of v2026. Key features: - Text-to-image generation with V6 model - Batch processing up to 100 images - Offline mode with cached models - Custom style presets and templates - Upscale up to 16K resolution - Prompt history and favorites Tech stack: Unknown Why it matters: This application provides a powerful tool for generating AI-generated images, ideal for both personal and professional use, including offline processing and batch image generation. Repository: amusedboot/MJ-Desktop---Midjourney-AI-Desktop-App-2026 GitHub stars: 8 GitHub description: MJ-Desktop - Midjourney AI image generator for Windows with offline mode and batch processing. Downloads: 5.8m+

### Overview: A versatile text-to-chart API that converts Graph Description Language (GDL) and simple text into images using various plotting engines like Graphviz, Gnuplot, Ditaa, and more. ### Key Features: - **Text-to-Image Conversion**: Converts GDL and simple text into images. - **Multiple Plot Engines**: Supports engines such as Graphviz, Gnuplot, Ditaa, and more. - **Web Editor Demo**: Available for quick testing. - **API Documentation**: Detailed documentation for API usage, including available engines and parameters. ### Tech Stack: PHP ### Why It Matters: This API is useful for developers looking to visualize data and concepts in a simple and intuitive way, making it easier to communicate complex information visually.

Overview: This project provides a Python library for efficiently transforming PDFs and Wikipedia pages into a Questions & Answers dataset, suitable for fine-tuning machine learning models. Key features: - Extract text from PDFs or Wikipedia pages - Remove references and introduction sections - Clean and split text into meaningful chunks - Generate questions and answers from text - Customizable text processing and dataset generation settings Tech stack: Python Why it matters: This library streamlines the process of creating a dataset from text files, making it easier to fine-tune models for various applications. Repository: antoninichiq/QADatasetBuilder Primary language: Python GitHub stars: 8 GitHub description: Efficiently Transform PDFs and Wikipedia Pages into a Questions & Answers Dataset for Fine-Tuning.

### Overview: Ozan is an open-source, low-slop model fine-tuned on a 12B-character long-form creative-writing LLM. It aims to produce prose that sounds more natural and avoids overused language patterns commonly found in large language models. ### Key Features: - **Low Slop**: Ozan is specifically designed to minimize slop, a common issue in large language models where phrases like "barely above a whisper" or "couldn't help but" are used. - **Craft Quality**: The model is evaluated on a comprehensive rubric, scoring highly in voice, imagery, coherence, and other aspects of writing quality. - **Training Data**: It is trained on curated, low-slop examples from strong non-GPT models and filtered human Reddit WritingPrompts, aiming to produce clean, readable prose. ### Tech Stack: The model is built using PyTorch and fine-tuned on Mistral-Nemo-Instruct-2407, with weights and sample stories hosted on Hugging Face. ### Why It Matters: Ozan aims to produce high-quality, low-slop creative writing that sounds more natural and avoids the overused language patterns commonly found in large language models. This makes it a valuable tool for writers looking to improve the readability and authenticity of their work.

### Overview: BERT (Bidirectional Encoder Representations from Transformers) fine-tuning is used to classify SMS spam/ham messages. The project compares Naive Bayes, BERT as a frozen feature extractor, and BERT fine-tuned on the same test set. Metrics include accuracy, precision, recall, and F1 score, with BERT fine-tuned achieving the highest accuracy and recall. ### Key Features: - Fine-tuning BERT for SMS spam/ham classification. - Benchmarking against Naive Bayes baseline and BERT as a frozen feature extractor. - Evaluated on Google Colab with verified evaluation metrics and confusion matrix. - Includes a detailed dataset split and evaluation metrics. ### Tech Stack: Python (Jupyter Notebook) ### Why It Matters: This project provides a comprehensive comparison of different classification methods for SMS spam/ham detection, highlighting the effectiveness of fine-tuning BERT for improved accuracy and recall.

### Overview: Qwen-Image is a production-ready RunPod serverless endpoint that provides a powerful text-to-image generation model with exceptional English and Chinese text rendering capabilities. This model is based on Alibaba's Qwen-Image, a 20B MMDiT image foundation model optimized for GPU performance. ### Key Features: - **Official Qwen-Image Model**: 20B MMDiT image foundation model. - **GPU Optimized**: Runs on A100 80GB, H100 PCIe, H100 HBM3, H100 NVL, and high-end workstation GPUs. - **Auto-scaling**: Scales to 0 when idle to save costs. - **Network Volume Storage**: Model cached persistently across all workers. - **Fast Cold Starts**: Optimized Docker image with pre-installed dependencies. ### Tech Stack: - Shell ### Why It Matters: Qwen-Image offers exceptional text rendering capabilities, making it ideal for applications requiring high-quality image generation with accurate text descriptions. The model's performance and scalability make it a valuable tool for developers and content creators looking to enhance their text-to-image projects.

Overview: Artlite is a lightweight AI runtime library designed for language models and image generation on AMD Ryzen AI processors, specifically the XDNA 2 NPU. It leverages the AMD XRT driver to run kernels directly, optimizing performance and energy efficiency. Key features: - Runs language models and Stable Diffusion on the XDNA 2 NPU. - Provides a small C++ library for runtime. - Supports prompt and generation on the NPU. - Offers tools like `artlite-cli`, `artlite-server`, and `artlite-bench`. - Includes a local model for coding agents. - Supports embeddings, rerank, and decision models. - Prefills the NPU with a local model for reduced energy consumption. Tech stack: C++ library, AMD XRT driver, oneDNN, LLVM, and xbyak. Why it matters: Artlite is a research project focused on pushing the boundaries of AI hardware, particularly the NPU, while optimizing energy efficiency and performance.

Overview: This repository contains recipes for running models on NVIDIA DGX Spark, focusing on LLM serving, image, video, and speech tasks. Each recipe includes the pinned image, engine version, and model revision, ensuring reproducibility and minimal host changes. Key features: - Recipes for LLM serving, image, video, and speech tasks - Runs on one or two DGX Spark nodes - Measured performance on each Spark - Includes model revision and engine version - Ships with the measured numbers - Runs without host changes: read-only model cache, no added privileges, output in `./out` Tech stack: Python Why it matters: These recipes provide a standardized way to run models on DGX Spark, ensuring consistent performance and ease of reproducibility across different environments. Repository: atakann/dgx-spark-recipes GitHub stars: 7 GitHub description: Measured recipes for the NVIDIA DGX Spark, on one Spark or two: LLM serving, image, video and speech, and vision (detection and tracking on video). Each one was run and measured on Sparks. Every recipe: pins the container image, the engine version and the model revision, ships with the numbers it was measured at, runs without host changes: read-only model cache, no added privileges, output in `./out`. Not affiliated with NVIDIA or any model maker. Each model keeps its own license.

### Overview: Godzilla llama.cpp is a fork of llama.cpp optimized for Windows with CUDA support, designed for specific hardware configurations like RTX 3080 and long-context workloads. It includes features like fused GDN paths, KV placement options, and speculative draft mechanisms. ### Key features: - **KDA / FlashKDA**: CUDA `gated_delta_net_cuda<…, KDA=true>` path; `LLM_ARCH_KIMI_LINEAR`. - **Fused GDN**: Fused GDN on by default for supported architectures. - **KV Surface**: Classic quants + turbo/TCQ/KVarN where compiled; `-ctk` / `-ctv`. - **Memory**: `--kv-ram`, `--kv-vram-only`, chunked prefill (`-ub`). - **Speculative**: DFlash, DSpark, native `draft-mtp`, CopySpec. - **Windows Build**: Documented MSVC + UCRT + CUDA pin. ### Tech stack: C++ and CUDA. ### Why it matters: This fork is optimized for specific hardware configurations and workloads, providing features like fused GDN paths, KV placement options, and speculative draft mechanisms. The default settings are based on lab measurements on a single box, not marketing SLAs. The features ship when they work, and those that fail are opt-in or off. ### GitHub stars: 8 ### Repository: atomicmilkshake/godzilla-llama.cpp

Overview: NeuroMem is a memory layer designed for AI assistants, providing persistent memory, semantic retrieval, and evaluation tools to ensure assistants remember users over time. Key features: - Memory storage - Smart retrieval - Chat integration - Provider flexibility - Evaluation suite - Dataset tooling Tech stack: Python Why it matters: NeuroMem offers a practical long-term memory stack for AI assistants, ensuring they remember users over time, improving the quality of their interactions and evaluations. GitHub stars: 8

### Overview Beatra Skills is a collection of AI agent skills designed to enhance creative production. These skills integrate seamlessly with Beatra, allowing users to generate, edit, and render AI-generated images, videos, music, and voiceovers directly on the Beatra platform. ### Key Features - **Image Generation**: Create high-quality AI-generated images. - **Voiceover**: Generate talking head videos. - **Music**: Generate AI-generated music. - **Video**: Edit and render videos. - **Text to Image**: Convert text into images. - **Image to Image**: Convert one image to another. - **Image Editing**: Edit and enhance images. ### Tech Stack The skills are built using Python and leverage the Beatra platform. ### Why It Matters These skills are essential tools for anyone looking to integrate AI-generated content into their creative projects. Whether you're a content creator, designer, or marketer, Beatra Skills can help streamline your workflow and elevate your projects to new heights. ### Repository Repository: beatra-ai/beatra-skills Primary Language: Python GitHub Stars: 8 GitHub Description: Agent Skills for Claude Code, Cursor, Codex, Windsurf, Cline, OpenClaw: AI image generation, text to speech, talking head video, AI music, and video generation skills that render on Beatra.