Mentor on Practical Data Science on the AWS Cloud Specialization mainly focused on SageMaker Pipelines (Data ingestion, Statistical Data Bias Detection, A/B Testing, and model deployment) and end-to-end Natural Language Processing models (BERT, FastText, BlazingText).

As the founder and organizer of this group, I organize talks with guests of the highest level. On the other hand, I give tutorial talks at different levels, teaching how to make a first model to complex pipelines with ensembles of models and neural networks.This community is an official part of the Google Machine Learning Communities.Link to one of my talks (How to get top 1% in Kaggle): https://www.youtube.com/watch?v=0o72ndnTnCQLink to join the group: https://www.meetup.com/es/TensorFlow-Santiago/Link to see the official google groups: https://www.tensorflow.org/community/groups

- Search Engine Optimization.- Building on Recommender systems and search engines at scale from scratch for Mercado Play

As a Data Scientist and Researcher of the group ICOVID Chile key indexes are generated to best represent the situation of the pandemic caused by the SARS-CoV-2 virus; this work is directly used by the Chilean Government and Decision Makers to fight the pandemic.

Led a team of 5 Data Scientists and developed and designed end-to-end time series forecasting and preventive maintenance models focused on the mining industry from tabular data or video/images.Provided translation between business needs to the analytics team and between the software developer team and the analytics team.

As the only Data Scientist in the company, I developed time series forecasting and predictive maintenance models and counterfactuals. The main focus is explainability and causality.Grew the company from scratch and managed multiples project in parallel.

Developed ping and network traffic time series forecasting models with uncertainty estimates and time-to-event models to predict the degradation of services and network congestion.

I was teaching the Deep Learning course with a practical approach using DNN, CNN, RNN, and transformers to solve real-life computer vision and natural language processing problems.

We introduced renewable energies to the problem, which introduces uncertainty in energy production and the model constraints. We proposed novel solutions using fictitious play-like heuristics and Deep Reinforcement Networks applied to equilibria in energy markets. In particular, a Deep Deterministic Policy Gradient (DDPG) algorithm is developed using state-of-the-art reinforced learning techniques and neural networks.

I proposed a bi-level model for European energy markets where the upper problem is a Nash equilibrium and the lower one is total cost minimization and solved for real-life types of cost functions and network size. Compared with what exists in the literature, it was between 10 to 100 faster while achieving the same mixed nash equilibrium strategies. The model and algorithm were developed in Julia.

Given historical customer data, I developed a demand forecasting model with uncertainty estimates using conformal predictions. It decreased forecasting errors in production by half compared to the previous one. I deployed the model using FastAPI and AWS Lambda.

I modeled a milk collection problem known in the literature as Vehicle Routing Problem with Time Windows plus custom constraints from our client. And solved using Big-M formulation plus heuristics using Julia programming language and the mathematical optimization modeling package JuMP for performance.The output is a spreadsheet with the optimal routes for each truck and relevant information for the company and driver.As a result, we decreased costs by 60% compared with the previous solution and reduced computation time by over 90%.

Given web-scrapped data from multiple job search webpages, I built a Natural Language Processing Model to identify duplicates. The output is an index to understand the country's economy by sectors, months, years, and cities. It is the evolution of an old system based on counting job posts from newspapers.

I improved a method that extracts a transcriptional regulatory network from a set of predicted transcription factors and binding sites by choosing the TFs and BSs most likely involved in the co-regulation by generalizing it and increasing precision. The work was done together with researchers at the Math modeling Center in Chile and based on their previous work, "Deciphering transcriptional regulations coordinating the response to environmental changes" https://doi.org/10.1186/s12859-016-0885-0

Developing an early cancer detection technique using differential expression of ovarian surface epithelium genes within a gene regulation network. I modeled it as a graph where its nodes represent the genes and the edges the gene expression regulation mechanisms in such a way that two genes are connected if one positively or negatively regulates the expression of the other Its dynamics are modeled in 2 ways:- Using differential equations that describe the concentration of the proteins and some metabolites.- Using Boolean variables (Kauffman Model)The transcription modules (a set of conditions and genes connected by a transcription factor) are studied to extract relevant biological information on the network. We use the Kauffman model and specific R algorithms for GRN to find them.The Institute of Biomedical Sciences (ICBM) from the University of Chile provided the data.