Chris Lee

This is an image extraction from a PDF utility module from one of my current projects. The project aims to extract table information from scanned pages of a reference book scanned into PDFs.

I needed to extract only the image data from the PDF to perform OCR on the scanned image. I used Indico's PDF extraction tool to extract the image data. I also have an image processing step to prepare for (optical character recognition) OCR. This involved applying thresholding to the pixel data, resulting in a higher contrast black-and-white image.

The rest of the project uses the Google Cloud Vision API to perform OCR on the extracted image. Then, it parses the resulting data using a combination of filters and pattern matching via Python regexes to determine the row and column of each cell. The resulting data is then written into an Excel file for the client.

Privvy is a system that synchronizes private files not tracked by Git in Git repositories using a more secure source. It makes use of Git hooks and actively sets the Git template directory to be used for all git repositories, but will not overwrite existing content (make a backup just in case).

It comes with Bash functions that are set up on package installation to either manually sync private files or to set up the Git hook integrations.

This was the first revision of an API framework written in Python using Tornado and Motor (MongoDB). It contains generic handlers for API routes as well as for user authorization.

It contains boilerplate that allows API routes to take shape quickly from scratch. It has a 100% test coverage run with a Travis integration.

A collaborative story-telling app. Users take turns adding short phrases to a story while only seeing a couple phrases prior.

This was a culmination of a semester course on Android development. It puts into practice various kinds of ItemAdaptors, modularized fragments, real-time database synchronization (Firebase), and Google User authentication and permissions.

Worked on portraying the ease of creating a new predictive model using Indico's custom collections API. It takes a couple of keywords, uses a Google image search to train any number of images (default=5), and can then predict new images as one of the provided keywords.

A lightweight web app that demonstrates the power of machine learning APIs. It is a React.js application that capitalizes on the reusability of components for a dynamic and clearly defined interface.

It also contains an API wrapper designed to allow easy access to the machine learning APIs.