At Konecto, I designed a closed-loop AI customer support system spanning knowledge ingestion, real-time voice delivery, and automated root-cause analysis. It supports 26 production voice agents across nine product lines and processes more than 1,000 daily calls; prompt and retrieval optimizations cut costs by 76% and search latency by 86%.
I also led delivery of a voice-agent quality-evaluation platform and built an enterprise PDF-to-RAG platform that orchestrates seven LLM providers. At Banco de Alimentos de Bogotá, I improved RAG relevance and latency, and developed NLP pipelines for large-scale text processing.
At Universidad de los Andes, I built the data-generation framework behind MARSA, a multi-accent Spanish anti-spoofing corpus, and designed a word-level partial-spoof pipeline. I also develop RAG systems for medical review analysis and teach graduate students NLP, retrieval, and agentic systems.
Spain
United Kingdom
Germany