At Conlan Scientific, I worked on SaaS playforms for numerous clients. This was where I had my taste of web development, and full stack web development in general. As a company, we specialised in designing sophisticated backend systems, with sleak, clean front end design. I was an essential developer for the projects that I worked on.
AutoLoanInsight(ALI) involved developing machine learned statistical models to predict risk in subprime loans using a range of data sources. All of this bundled in a clean and flexible SaaS platform, to be able to easily allow the non-technical client the ability to trivially train new models with ever updating data. The techinicalites of this project includes a intricate full stack Django design, query optimisations, statistical analysis using standard python packages and financial institution savviness.
Adoptimize had a simple mission statement, make it easier for animal shelters to take good looking pictures of the animals in need
of adoption. This involves shelters plugging in a webcam, and recording and uploading in realish time to the platform.
The video files were then put through an elaborate pipeline in order to generate "instagramable" images of the dogs automatically. This project was a massive adventure in Computer Vision, deep learning, machine learning and statistics. Background subtraction, super resolution and neural networks were just the beginging. Even IO considerations had to be accounted for. The whole project was a small startup, so costs had to be kept minimal. This meant that despite the monumental amount of processing that needed to be done, it all had to be done on low end hardware.
On top of all the technical difficulties on the backend, I learnt a lot on the front end web design. The interface was crammed with flexibility, without ever being cluttered and always intuitive. When mixing a lot of technologies together, it was really valuable to be the responsible for the entire stack, as it allowed us to always find the cleanest solution. This did mean that the front end and backend were tightly coupled, but that is what needed to happen in order for the product to be a success.
The third major project was developing and online learning platform. While it was likely the “simplest” of the major projects, it still involved elements of natural language learning, in order to automatically generate keywords from articles, and create a dynamic living index of keywords among all articles. While not nessasiry a revolutionary construct, it brought together the worlds of language and design.
Other large pieces of work included siginificant database cleaning ETL jobs with very unclean and sometimes unstructured data. Wrapped up into a very nice package with working interface. We did a lot of very clever tricks to integrate into an exisiting database, and created a very dynamic and flexible script that will be future proof for a long time to come.
In my final few months at Conlan Scientfic I helped a groupd trying to model influence of political figures/lobby groups. This was a rare chance to flex my EE muscles, when I decided that it models quite nicely as an electrical circuit.