Resume

The goal was to optimize build + testing phase, which took long time and caused a lot of distruptions. What started as long research phase of existing systems turned into custom Build System based on Docker and Go.

While early versions allowed to find inconsistencies, project was still WIP in my final days.

Overarching, multi goal project on the topic of tangling heterogenous codebases, code generation and logic discovery.

This project involved researching, for feasability various existing solutions and writing various utility as a proof of concept in a few lesser known languages.

This project started as an answer to “how update stack not updated over the long period”. I was working as a lone developer and acted as a time progressed in intervals, slowly updating whole stack.

Project took couple of months but no downtime happened and whole update was relatively safe.

Page renders were known to be slower than expected, but multiple sweeps of optimizations weren’t able to produce satisfying results.

Upon hooking extra telemetry I found out that the cause was a single table, surpisingly much bigger than the development model which was missing index.

Time-to-render decreased by ~1.5 DOMs.

Task was to have a virtual roof builder, that would both connect compatibility rules but also look nice. Due to miscommunication with vendor, I volunteered and was tasked to integrate provided model with the data before the marketing event set for the promotion of the feature.

Builder was delivered on time.

Goal of the system was to provide smart notepad for medical staff so that they can focus on work instead of clicking on fields.

Project failed after initial phase due to lack of funding.

Concept for the runner-kind of game with economic layer targeted at young girls to help them make economic decision in the face of omnipresent microtransactions.

Core game was gaining and spending currency while teaching that impulse spending hampers progress.

Project didn’t reach development phase due to lack of financing.

Savegame Pipeline was a first Elixir project that I advocated after we’ve noticed unreliability when Ruby-based system connected to Heroku’s database. Feature, while sounding low-priority was crucial for operations, as savegames were used in math students performance evaluation.

Pipeline was running on it’s own, without updates, for years before it was shutdown.

Corporate partner wanted to have, what is today known as, content farm. These weren’t popular back then and the heart of solution was to ensure that there sites are visually different and content had differences enough significant so that they wouldn’t decrease search ranking.

I architected solution based on special syntax that would allow variants of parts of text and included custom editor that allowed to check specific version of it.

Implementation was done in Python/Django but my role was limited to guidance during unclear situations.

Project itself wasn’t complicated, however 2 weeks before the launch during manual tests what was found is that search engine is inconsistent and has tendency to bleed sesssions.

I was appointed emergency-leader to solve the issue. After analyzing I made decision to completely rewrite search engine, which resulted in getting on time and only 2 days of overtime.

One of our partners produced gigabytes of real estate offers daily. Most of those were very similar, but differed in branding. Because not all export consumers took incremental packs, big ones were produced, causing almost 24h processing loop.

This project was bootstrapped by myself, both in terms of architecture and documentation for approval by the partner. In gist it was about moving each offer in a multiple “processing belts” and caching results, so one image wouldn’t be processed unless changed, most of the data pack could be cached.

Implementation ran the whole process in less than 15 minutes and allowed our partner to produce even more variation of the offers, with minimal impact on the processing time.