Applied AI · Document AI / OCR
I build applied systems for documents and vision: spatial OCR, detector orchestration, and structured extraction. The model orchestrates; deterministic code decides. Background in pharma and retail operations (SKU, lots, invoices, pharmacy channel).
Site: javi2481.github.io · LinkedIn: javier-berrone
Conversational replenishment over a structured catalog. The LLM interprets; Python owns the order quantity. Goldens in CI.
PaddleX detectors (objects, faces, pose, vehicles) over a single photo. Toggle layers and see what each adds.
Academic OCR SPA with PP-OCRv6: spatial boxes, edit, export JSON / Markdown / CSV / annotated PNG.
Claims intelligence kernel. Shipped instance: BYMA financial statements. Typed claims are the source of truth; RAG chat is optional.
Local demo: PDF or image to Markdown with VLMs via Hugging Face Inference Providers, plus A/B comparison.
Data analyst at a pharmaceutical wholesaler (2020–2023): operational and pharmacy-channel data. Independent consultant since 2024, shipping document-AI and computer-vision prototypes (repos above).
Completing the Tecnicatura Superior en Ciencia de Datos e Inteligencia Artificial at ISTEA (official degree, Argentina). Expected graduation: December 2026.
Python · Docker · FastAPI · TypeScript · Computer Vision · OCR / Document AI
Applied AI · Document AI / OCR. OCR espacial, orquestación de detectores, extracción estructurada. El modelo orquesta; el código decide. Background en pharma y retail (droguería, SKU, lotes, canal farmacias). ISTEA, egreso diciembre 2026.
Sitio: javi2481.github.io