Projects

A lightweight, production-ready NPM package for generating text embeddings locally using @xenova/transformers. Supports MiniLM and BGE models with 384-dimensional vectors. Features include batch processing, similarity functions (cosine, euclidean, dot product), and built-in benchmarking tools. No external API calls required - runs entirely in Node.

High-performance, API-driven vector database designed for fast and meaningful semantic search. Vectra externalizes embedding generation and vector similarity search from core applications, using a local embedding model and HNSW indexing to deliver context-aware results that outperform traditional rule-based and SQL-style querying on large datasets.

Elentis is a minimal, folder-based REST API toolkit for building clean and organized APIs on the Deno runtime. It leverages Deno's native TypeScript support and modern features to provide a lightweight framework that emphasizes simplicity and developer experience. With Elentis, developers can quickly set up RESTful endpoints using a file-system-based routing approach, making it easy to manage and scale APIs without the overhead of traditional frameworks.

CHME, Compact Hierarchical Memory Engine (CHME), is an in-memory memory orchestration engine that provides multi-collection support, keyword-based retrieval, automatic routing, and snapshot persistence for LLM applications. It features deterministic behavior and supports both local (Ollama) and cloud (OpenAI-compatible) providers. Made with curiosity, not complexity. It may not have achieved significant success, but it made sense as a concept. The project also includes a technical paper outlining the architectural approach, along with benchmark results and detailed technical infrastructure documentation.