Luis Miguel Benitez Ruiz

I collaborated in building a data analysis pipeline that orchestrated multiple services to annotate items in Amazon's apparel catalog automatically. I then automated predictions using MxNet networks and extracted meaningful attributes from catalog images (e.g., floral pattern, scoop neck), which were used to enhance the catalog items, so Amazon customers can use them to search or filter when shopping for clothing.

I was involved in building the prediction services that used MxNet, and a fast-annotation and review service: a web application so data annotators could review low-confidence predictions. Also built a data warehouse to collect metrics throughout the whole process. This was used by our data scientists to improve the existing models and perform new experiments.

I developed an on-call rotation monitoring tool while working as part of a DevOps team. This app monitored our bug-tracking databases and allowed engineers to create on-call shifts for their chosen time frames and products, set up alerts to their email and mobile number, and allow managers to set up escalation rules for high-severity incidents.

Before this, the organization had no tool for managing on-call, so rotation was controlled differently across teams, mostly through shared documents or emails. Setting up alerts was complicated and required engineers to have privileges in systems they usually should not have had access to.

A centralized tool for on-call gave better visibility and accountability to the on-call rotation for all teams, made the onboarding process easy and quick for new team members, and helped keep sensitive systems more secure.

It was so well received that what began as a personal project for my team quickly became a new internal product, part of the standard tools for the entire organization.

Collaborated in developing a web application to allow the fast manual labeling of very large data sets through the use of online machine learning.

The application iteratively built classification models and generated quality metrics as the user provided labeled data. The annotator had a very streamlined and simple UX, and data scientists had access to aggregated metrics and other metadata for the generated models.

With this tool, annotators had to label only from 1% to 4% of the samples in a dataset, while before, they had to manually label 100% of the dataset. This allowed a much quicker turnaround when generating good-quality labeled datasets for our data science team.

CloudFix is a cost-optimization service for AWS. It scans AWS resources in an organization, identifies inefficient resources, and automatically and securely fixes them.

I collaborated in designing the architecture of the CloudFix framework, designing the detection and fixing algorithms, and implementing the product in TypeScript and Node.js.