Edson Cilos Vargas Júnior

This is an independent project. It is an MCP-native knowledge platform on PostgreSQL that exposes a semantic artifact store to LLM agents. Retrieval combines pgvector dense search with PostgreSQL full-text search, fused through Reciprocal Rank Fusion over model-versioned embeddings. The platform tracks provenance across artifacts and includes a signal analyzer that detects gaps in projects and other AI systems.

This is an independent project. I built a native SwiftUI app for Mac, iPhone, and iPad that brings speech recognition, multiple LLM providers, coding agents, terminals, and Git workflows into one interface. Integrations run over JSON-RPC and NDJSON streams, with session resumption, interruption, and failure handling.

I designed and published a Portuguese-language applied machine learning course: 19.5 hours across 109 lectures, built around Python and scikit-learn, covering mathematical foundations and end-to-end projects. It has 750+ learners and is rated 4.6/5 based on 109 ratings.

I implemented Principal Component Analysis, Linear Discriminant Analysis, and Laplacian Discriminant Analysis (with variants) from scratch for facial recognition, including low-level routines such as eigenvalue computation. I wrote in-depth documentation of each method so the client could build in-house expertise in dimensionality reduction for computer vision. I also grounded the implementation and the project's main findings in a review of the scientific literature.