At ARY S.A.S, I contributed to the design and development of voice-based conversational agents, integrating real-time speech recognition and synthesis. I also designed RAG architectures and conversational memory components so agents could use client knowledge bases and maintain context across long conversations.
I applied knowledge distillation, quantization, and WebLLM/WebGPU optimization to enable lightweight client-side inference, and developed a compact ONNX lip-sync model through distillation from MuseTalk. At Universidad Nacional de Colombia, I contributed to web platforms for Saber Pro pre-registration and academic program self-assessment, and designed database schemas for information systems.

