Enrico Z. Borba

A complete redesign of the FOSSA Issues API and database schema, leading to around 120 times speedup on SQL queries. FOSSA is a SaaS that offers legal and security auditing tools for code bases. The workflow is roughly as follows:
• Scan your code.
• Review issues.
• Resolve them.
• Re-scan your code.
• Repeat.

When I joined FOSSA, large code bases would result in tens of thousands of issues being reported. The original database representation of issues led to extremely slow and complex queries being required for even the most simple of pages on the app. For example, downloading all of its issues across its code bases would take hours for a large organization, even though the resulting CSV file generated would be only around 200MB.

I designed a new API and database schema for issues leading to:
• Simpler SQL queries, leading to better developer experience
• Simpler API, leading to better developer and customer experience
• Faster queries

I designed the schema and API so that both the old and the new schema and API could be used simultaneously to allow for gradual roll-out. Additionally, the migration I designed was completed with zero downtime, and in a matter of hours, despite migrating over 10 billion rows.

A complete redesign of the FOSSA Issues UI and UX with complex user interactions, such as filtering, rich searching, and triaging of issues.

When I joined FOSSA, large code bases would result in tens of thousands of issues being reported. This resulted in a poor user experience as loading issues were slow, searching was brittle, and filtering did not exist.

I led a team of three engineers and a designer to rework the UI to allow for rich searches, filtering by a variety of fields, and clear triaging of issues. This was the most successful project at FOSSA in that:
• There was no scope reduction.
• It was delivered on time within six months.
• It was executed so well that it could be rolled out smoothly.

The UI rework that I oversaw and the back-end changes that I designed and implemented led to a speedup of 120 times on issue page loads for the largest organizations.

A website for sharing templates for the reMarkable tablet.

The design is inspired by the operating system's interface on the reMarkable. The front end is built with React and TypeScript, and the back end is a simple use of FastAPI on Python, which interfaces with Google Storage, and is deployed on Heroku.

A Rust crate for quickly building component-based text-based user interfaces, i.e., tangible user interfaces (TUIs). React and SwiftUI heavily inspire it, and it contains features that resemble functional components, hooks, and a (mostly) declarative DSL.

A chemical reaction network simulator in Python that is used to facilitate some homework exercises in Caltech's Bi 191ab, a biomolecular computation course.

This package takes stoichiometric equations as input along with reaction rates and initial concentrations and outputs final concentrations along with a plot of concentration of all the species over time. This package uses Python 3 operator overloading to create a pseudo-DSL to resemble stoichiometric equations for more readability.

A UI to simplify human-in-the-loop system repairing of segmented cell division video data. The tool that I wrote was eventually patented because of its novelty and ease of use.

Cell division tracking is an intensive process of manually reviewing frame-by-frame microscope video data of cells and annotating the lineage of the cells across frames. At the Van Valen Lab, we trained a computer vision model to segment and track cells without the need for a human. However, during the initial training of the model, it would often incorrectly track cells. The tracked cell data was initially represented as a mixture of TIFFs and CSVs, with the TIFFs containing the segmentation information and the CSVs containing the lineage information. This representation made is very time-consuming and unintuitive to fix incorrect tracking data.

I wrote a tool to let scientists visually correct tracking data by overlaying colors onto the cell video, clearly identifying the segments and lineages. This simplified and sped up the process of repairing incorrect data, allowing the team to train the models better. The tool was initially written in Python and then ported to the browser using JavaScript.