Daniel Hoyos

Background

About Daniel Hoyos

I'm a Lead Machine Learning Engineer with over seven years of hands-on experience, specializing in LLMs, NLP, time series forecasting, and ML-driven Predictive Analytics.Past Experience: At Blue Orange Digital (NY-based), Developing Copilot For Insights In Partnership with Tungsten Automation, an LLM-based product optimized for enterprise-grade document ingestion and analysis through a conversational interface. In past projects, I've used NLP and GenAI for document optimization, notably for a leading vehicle company utilizing open and proprietary LLMs; I have also done predictive analytics in the marketing space for one of the biggest entertainment companies in the US.Past Experience: At MercadoLibre, Latin America's e-commerce giant, I innovated AWS spot instance forecasting, enhancing infrastructure decision-making and cost-efficiency.Specialties:NLP and LLMs: Developed dialogue systems, OCR, and ASR frameworks. Leveraged text mining and clustering for user insights, emphasizing the Pareto principle. Proficient with models like Transformer, BERT, Roberta, GPT series, and Llama Series, primarily using Hugging Face for open-source models; I've been using them since 2018. Finally, tons of work in the last year and a half with proprietary models from OpenAI and Anthropic. Time Series Forecasting: Tree-based and NN-based forecasting models. The opensource Temporal Fusion, Transformer Pytorch version, is currently being used in the package Pytorch Forecasting.Marketing ML: Devised strategies for predicting churn, attrition, and customer lifetime value, facilitating targeted marketing campaigns.Deep Reinforcement Learning: Applied advanced techniques like DQN, DDPG, REINFORCE, and PPO for real-world continuous state-action scenarios.