Projects

Overview: This repository contains code for fine-tuning Transformers for sentiment analysis and transitioning from research to production with containerized training jobs. Key features: - Sentiment Analysis Fine-Tuning - Containerized Training Jobs Tech stack: Jupyter Notebook Why it matters: This project showcases the practical application of Transformer fine-tuning for sentiment analysis and provides a guide on transitioning research projects to production using containerized training jobs.

Overview: OmniChunk is a library designed to intelligently split files into context-rich chunks, preserving semantic boundaries while maintaining traceability and semantic integrity. It supports 15+ programming languages, deterministic output, and zero external dependencies, making it perfect for RAG systems, code analysis, and LLM context optimization. Key features: - Supports multiple programming languages - Preserves semantic boundaries - Provides deterministic output - Integrates with 15+ languages - No external dependencies - Perfect for RAG systems, code analysis, and LLM context optimization Tech stack: Python Why it matters: OmniChunk is a powerful tool for structuring and analyzing code and documents, ensuring that the integrity and traceability of the original content are maintained while allowing for efficient processing and integration with downstream systems.

Overview: This repository is a comprehensive AI bootcamp for developers, focusing on NVIDIA Inference Microservices (NIM) microservices. Participants learn to build real-world applications, set up Docker containers, and utilize REST API endpoints for inference requests. Key features: - Building RAG via NVIDIA NIM APIs - Building RAG with a Localized NVIDIA NIM - Running NVIDIA NIM with LoRA Adapters - Training own adapters on custom datasets using single-GPU and multi-GPU strategies - Multimodal NIM (VLM) - Utilizing NVIDIA NIM as Agents - Deploying NVIDIA NIM Blueprints Tech stack: Jupyter Notebook Why it matters: This bootcamp provides a hands-on overview of deploying NVIDIA NIM Blueprints, fine-tuning models using Parameter Efficient Fine-Tuning (PEFT) techniques, and utilizing multimodal NIM, making it a valuable resource for developers looking to enhance their AI skills in the field of generative AI applications.

Overview: 这个仓库包含44个可以直接套用的电商AI出图中文Prompt模板，适用于商品场景图、服装模特试穿、配饰与姿势、带货短视频分镜等场景。模板库中的内容无需修改，直接使用即可。 Key features: - 商品场景图 - 服装模特与试穿 - 配饰与姿势变化 - 带货短视频分镜 Tech stack: PixGT Why it matters: 这个库为电商出图提供了标准化的Prompt模板，确保了出图的准确性和一致性，提高了转化率。

Overview: Build an AI-driven portfolio app using Next.js and Tailwind CSS, exploring advanced AI technologies such as vector embeddings and vector databases. This project is ideal for showcasing your skills to recruiters. Key features: - Vector embeddings and vector databases integration - Next.js and Tailwind CSS implementation - Open AI API integration Tech stack: JavaScript Why it matters: This project demonstrates your ability to implement cutting-edge AI technologies and showcase your skills effectively to recruiters.

Overview: This project aims to establish a comprehensive framework for evaluating and analyzing Large Language Models (LLMs) through structured intelligence reports. Unlike typical AI research papers or commercial reports, it provides detailed insights into AI model performance, benchmarking methodologies, multi-hosting provider analysis, and industry trends. Key features: - Benchmarking methodologies - Multi-hosting provider analysis - Industry trends Tech stack: Python, Markdown, Git, and various LLMs and hosting providers. Why it matters: This framework offers a structured approach to understanding and comparing LLMs, providing valuable insights for AI decision-makers and researchers. Repository: rawalraj022/aiprl-llm-intelligence-report GitHub stars: 10 GitHub description: This repo establishes (AIPRL-LIR) framework for evaluating and analyzing LLMs through structured intelligence reports. Unlike typical AI research papers or commercial reports, it provides insights into AI model performance, benchmarking methodologies, multi-hosting provider analysis, and industry trends. README excerpt: <div align="center"> # 🌍 (AIPRL-LIR) AI Parivartan Research Lab - LLMs Intelligence Reports ### *Leading Models & Companies, 23 Benchmarks in 6 Categories, Global Hosting Providers, & Research Highlights* [](LICENSE) []() [![AI Research](https

Overview: This project showcases how to build real-time inference pipelines using Ray Serve, focusing on the deployment graph API for scalable and modern ML deployments. Key features: - Deploying deep learning models with Ray Serve - Using the deployment graph API for efficient resource allocation - Implementing business logic with Ray Serve - Building multilingual chat applications with Ray Serve Tech stack: Jupyter Notebook Why it matters: This tutorial provides practical guidance on leveraging Ray Serve for building scalable and efficient real-time inference pipelines, making it valuable for ML practitioners and engineers looking to modernize their ML workflows.

### Overview: Latent Memory is a Module for Large Language Models designed to integrate a vector-based memory system into the inference process, leveraging embeddings to capture deeper semantic meaning. ### Key Features: - **Vector-Based Memory**: Stores conversations as embeddings in a vector database, enabling nuanced semantic preservation. - **Context-Aware Inference**: Seamlessly integrates conversations into the LLM's inference process, enhancing the model's ability to understand context and nuances. - **Efficient Storage**: Uses various formats (CSV+JSON, NumPy+JSON, HDF5, Pickle) to store embeddings and metadata, ensuring scalability and security. ### Tech Stack: Python ### Why It Matters: This project aims to improve the performance and effectiveness of LLMs by integrating a more sophisticated memory system, potentially leading to better understanding and context-aware responses.

### Overview: This project is an intelligent image gallery application powered by AI, utilizing Qdrant for vector database search and deduplication. It supports text-based image search and uploads, making it a versatile tool for managing and searching through your digital photos. ### Key features: - Text-based image search - Image uploads - Image deduplication ### Tech stack: The project is built using Python, Qdrant for vector database, Gradio for user interface, and SentenceTransformers for image processing. ### Why it matters: This application streamlines the process of managing and searching through digital photos, making it an invaluable tool for anyone looking to organize and find their images efficiently. The integration of AI and vector databases ensures that the search functionality is both powerful and user-friendly.

Overview: Steadlith is a Python library designed for engineers managing a single-host RAG index, reusing unchanged RAG chunks with content-defined identities, cache-aware planning, and transactional indexing. Key features: - Reuses unchanged RAG chunks with content-defined identities - Cache-aware planning - Transactional indexing Tech stack: Python Why it matters: Steadlith simplifies the management and reuse of RAG chunks, ensuring efficient and consistent indexing while reducing the overhead of retraining and embedding updates. Repository: satwiksps/steadlith GitHub stars: 10 GitHub description: Steadlith reuses unchanged RAG chunks with content-defined identities, cache-aware planning, and transactional indexing. README excerpt: <div align="center"> <h1 align="center">Steadlith</h1> <img src="https://raw.githubusercontent.com/satwiksps/steadlith/main/assets/steadlith-banner.svg" alt="Steadlith" width="100%"> <p>Steadlith reuses unchanged RAG chunks with content-defined identities, cache-aware planning, and transactional indexing.</p> <p> <a href="https://steadlith.readthedocs.io/en/latest/">Documentation</a> &middot; <a href="https://steadlith.readthedocs.io/en/latest/getting-started/quickstart/">Quick start</a> &middot; <a href="https://steadl

### Overview: Lancelot is a Ruby library that provides a native, idiomatic interface to the Lance columnar data format, enabling high-performance vector search, full-text search, and hybrid retrieval in Ruby applications. ### Key Features: - **Dataset Creation**: Create Lance datasets with schemas. - **Data Storage**: Add documents to datasets. - **Document Retrieval**: Read documents from datasets with enumerable support. - **Vector Search**: Create vector indices and perform similarity search. - **Full-Text Search**: Built-in full-text search with inverted indices. - **Hybrid Search**: Combine text and vector search with Reciprocal Rank Fusion (RRF). - **Schema Support**: Define schemas with string, float32, and vector types. - **Row Counting**: Get the number of rows in a dataset. ### Tech Stack: Lancelot is built on top of the Lance Rust crate, providing a Ruby-native interface to Lance, a modern columnar data format for ML. ### Why it Matters: Lancelot simplifies the process of working with large datasets, enabling efficient storage and search of multimodal data including text, vectors, and more, making it ideal for applications requiring high performance and scalability.

### Overview: TensorVillage is a powerful tool designed to organize and manage AI models scattered across multiple drives. It seamlessly integrates your models into a single, organized drive, making them accessible and easy to manage. ### Key Features: 1. **Unified Drive**: All your AI models are stored in one organized drive, making them easily accessible and manageable. 2. **No Copying or Linking**: Models are read directly from their original locations, ensuring they remain intact and unaltered. 3. **Safe Scanning**: It reads headers rather than files, ensuring that terabytes of data can be scanned in seconds. 4. **Sorting Algorithms**: It uses sorting algorithms to organize models based on their names, shapes, and metadata, making it sortable and safe. 5. **No Unpickling**: It avoids unpickling pickles, which can execute code, ensuring models remain intact and secure. 6. **Real Filesystem**: Built on WinFsp, it serves files directly from the drive, allowing for efficient and secure storage. ### Tech Stack: - **WinFsp**: The driver behind the drive, enabling efficient and secure storage of AI models. ### Why It Matters: - **No Copies or Links**: Ensures that your models remain intact and unaltered. - **Safe Scanning**: Reads headers rather than files, ensuring efficient and safe scanning of terabytes of data. - **Sorting Algorithms**: Uses sorting algorithms to organize models, making them easily accessible and manageable. - **Real Filesystem**: Built on WinFsp, it serves files directly from the drive,

Overview: This project involves creating a highly scalable Qdrant vector database using Go, leveraging Terraform for infrastructure as code, and integrating with AWS and Kubernetes for deployment and monitoring. Key features: - Automated snapshotting and backup options - Recovery mechanism from backup - Efficient ingestion of 1 million records - Observability and performance monitoring with alerts - Use of Terraform for infrastructure - Deployed using Docker Compose for local and production environments Tech stack: Go, Go Fiber, Prometheus, Grafana, Qdrant, Terraform, AWS, Kubernetes, Apache Kafka, Trivy Why it matters: This project showcases the ability to create a scalable and reliable vector database solution, integrating with cloud providers and monitoring tools, which is crucial in modern DevOps environments.

Overview: Mind-X is an intelligent assistant designed to understand and assist with your daily tasks, evolving in real-time to become a next-generation PersonAI system. Key features: - Real-time task assistance - Personalized learning and improvement Tech stack: Go Why it matters: Mind-X is a cutting-edge AI system that enhances productivity and personal efficiency, making your life easier and more organized. Repository: sjy-dv/mind-x Primary language: Go GitHub stars: 10

### Overview: Kimari Local AI is an open-source framework designed to run powerful language models on older NVIDIA GPUs, specifically the GTX 1060 and GTX 1080. It includes a CLI-first workflow, GPU-aware profiles, llama.cpp / GGUF runtime support, an OpenAI-compatible local endpoint, a Gateway Dashboard, and local integration helpers for various tools. ### Key Features: - **CLI-first Workflow**: Simplifies installation and management. - **GPU-Aware Profiles**: Optimized for older GPUs. - **llama.cpp / GGUF Runtime Support**: Enables running LLMs on consumer hardware. - **OpenAI-Compatible Endpoint**: Compatible with existing tools and integrations. - **Gateway Dashboard**: Provides a local dashboard for monitoring and management. - **Integration Helpers**: Supports Open WebUI, Continue.dev, OpenClaw, and Hermes. ### Tech Stack: - Python 3.10+ - llama.cpp - OpenAI-compatible API ### Why It Matters: - **Older GPU Support**: Designed specifically for GTX 1060 and GTX 1080, not just the latest cards. - **Zero Cloud Dependency**: Everything runs locally, no subscriptions, no API keys, no telemetry. - **OpenAI-Compatible Endpoint**: Drop-in endpoint for existing tools and integrations. - **CLI-first**: One command to install, one command to start, one command to diagnose. - **Honest Status**: No inflated benchmarks, no "coming soon" claims. Alpha means alpha. ### Current Status: - **Framework / CLI**:

Overview: HFTrie is a dynamic data structure designed for efficient indexing and querying of 64-bit binary vectors, offering fast range search capabilities. Key features: - Dynamic indexing for efficient addition, deletion, and range search. - Efficient use of multiple bits per layer to increase fanout and reduce time complexity. - Performance metrics for various index sizes and radius values. Tech stack: C++ Why it matters: HFTrie provides a fast and accurate solution for binary vector indexing and search, making it suitable for applications requiring high-speed nearest neighbor queries.

Overview: Quotify-AI is an AI-powered quote generator bot that posts transformative quotes and beautiful illustrations on Instagram, with the potential to expand to other social media platforms. Key features: - AI-generated quotes and illustrations - Integration with vector database and LLM API - Support for Instagram - No support for Reddit, Facebook, or Twitter Tech stack: Python Why it matters: Quotify-AI leverages AI to create engaging and visually appealing social media content, making it a valuable tool for anyone looking to boost their brand's presence on Instagram.

Overview: `rag-doc-qa` is a production-style Retrieval-Augmented Generation (RAG) document Q&A system that includes LangChain ingestion, hybrid retrieval, citation-grounded answers, and a streaming chat API. It uses a combination of sparse BM25 and dense embeddings, with a cross-encoder rerank and a citation-grounded approach. Key features: - Pluggable ingestion for various document formats - Hybrid retrieval with BM25 and dense embeddings - Two vector stores: Chroma for local development and pgvector for production Postgres deployments - Citation grounding for every answer token - Streaming chat API with FastAPI and Server-Sent Events (SSE) - Ragas evaluation for context precision, recall, and answer relevancy - Config-driven setup with YAML profiles and environment overrides Tech stack: Python 3.11 Why it matters: This system provides a comprehensive solution for document Q&A, addressing the limitations of traditional RAG demos by incorporating hybrid retrieval, citation grounding, and a streaming service that handles high concurrency. Repository: `tainguyen07/rag-doc-qa` GitHub stars: 10 GitHub description: Production-style Retrieval-Augmented Generation document Q&A: LangChain ingestion, hybrid retrieval, citation-grounded answers, streaming chat API.

Overview: This project integrates OpenViking's knowledge base with nanobot, enhancing semantic search, RAG context augmentation, and document indexing capabilities. It utilizes FastAPI and Python for a robust and efficient integration. Key features: - RAG Augmentation - Semantic Search API - Resource Management - File Browser API - CLI Tool - Web UI Compatible Tech stack: Python Why it matters: This project provides a powerful tool for developers looking to improve the quality and relevance of their nanobot's responses, making it easier to access and manage knowledge across various resources. Repository: tankyhsu/nanobot-viking GitHub stars: 10 GitHub description: OpenViking knowledge base integration for nanobot — RAG augmentation, semantic search, vector embeddings, document indexing. Free models on SiliconFlow.

### Overview: A-MEM Agentic Memory System is a memory management tool designed for LLM agents based on the Zettelkasten principle. It integrates seamlessly with IDEs like VSCode, providing features such as automatic note construction, link generation, memory evolution, and semantic retrieval. ### Key Features: - **Core Features**: - Automatic extraction of keywords, tags, and contextual summaries - Automatic linking of similar memories - Dynamic updating of existing memories - Intelligent search with graph traversal - Configuration via `.env` file - Support for multiple providers (local Ollama, cloud OpenRouter) - Parameter validation - Safe graph wrapper for robust operations - **Advanced Features (New)**: - Automatic classification of notes into 6 types - On-the-fly priority scoring based on type, age, usage, and edge count - Background processes for graph maintenance (Link Pruner, Relation Suggester, Dead-End Node Auto-Linker, Summary Digester) - Automatic scheduler for memory enzymes - Experimental fields support without schema changes - Support for local Docker-based Jina Reader instance - Unstructured PDF extraction using Unstructured library or API ### Tech Stack: Python ### Why It Matters: This implementation provides a robust and efficient memory management system for LLM agents, integrating seamlessly with IDEs and supporting multiple LLM providers. It offers advanced features like automatic classification and priority scoring, making it a valuable tool for developers working with large datasets and complex models