I'm Keylor, a software developer from Costa Rica. I like building complete products: the database, the API, the interface, and the tests and CI that keep everything working.

Right now most of my time goes into HealthyCropping, an Android app that recognizes coffee, cacao and banana diseases from a photo without an internet connection. Before that I built Booth, a catalog and ordering tool for small shops that sell through WhatsApp, and a RAG engine that answers questions using only the documents you give it.

Outside of the code I'm studying cloud and security, which is where the certifications below come from.

Point a phone at a coffee leaf, a cacao pod or a banana leaf and get a diagnosis right there in the plot. Each crop has its own MobileNetV3 model running on the device through TensorFlow Lite, and scans, farms and plots are stored locally first. Next up: sync and a Gemini RAG layer that explains what to do and cites official agronomy manuals.

Small businesses already sell through WhatsApp, one chat at a time. Booth gives each store a single link with its catalog, turns the cart into a clean WhatsApp message, reserves stock atomically and uses Gemini to read transfer receipts, so owners know what to check before confirming a payment.

Upload PDFs and notes, ask in English or Spanish, and get answers that cite the exact passages they use. Retrieval fuses pgvector and Postgres full-text search with Reciprocal Rank Fusion, an optional Cohere reranker reorders the results, and a calibrated gate refuses before the model is called when the documents don't hold the answer. An Angular web app shows each answer next to its sources, and a benchmark runs on every pull request and fails if quality drops.

Pulled from the GitHub API and refreshed every day.

_{Thanks for stopping by.}