Flexible Demeanour

Personal site of Alex Bachurin

Background in Mathematics. Used to be Investment Banker, switched to Programming several years ago. Can do anything in Python, a lot in React Typescript, occupy any role in Web Development - Frontend, Backend or Full Stack.

What's so funny?

Technology

Python, including FastAPI, Dash
React Typescript
HTML, CSS, vanilla Javascript
MongoDB, SQL
ML/RL

full cv

Preferred environment

I work on Mac, but also have a lot of experience dealing with Microsoft products like Excel, Sharepoint etc.

I had a lot of freelance projects of various nature and scope for different clients. All of them were mostly solo projects. Click below for more details.

RL 2048

Big Dash App

Miscellaneous

E-mail

* I live in the south of Portugal, GMT+1

My personal pet project

I actually learned to code in order to make the first version several years ago. Since then I use it as a training ground for any newly acquired skills.

At first existed just as a Reinforcement Learning project, later I made a User Interface for it, eventually turned it into a Web App.

Reinforcement Learning is applied to a well-known smartphone game. Algorithm knows only current score. Starts with random moves and quickly (in minutes!) learns to beat an average human. I invented a training method myself ... and later discovered that, sure enough, it was already known 😊

Current iteration = React Typescript as Frontend + FastAPI as Backend + Python Worker + MongoDB Database.

Used to host it on AWS EC2, then AWS BeanStalk, now on DigitalOcean App platform.

First version frontend was built with Dash Python

Now rewriting the main calculation engine in Julia.

Detailed description, as well as Github links, can be found at the "HELP" page of the Project:

RL Robot 2048

My most recent and the biggest project so far

A complex multi-user App with a lot of background data collection and processing.

There was a lot of legacy code, ~10000 lines, Python + CSS. Initially I've been asked to review/improve/optimize it. Had to rewrite everything in the end. Took ~1 year to complete.

Final configuration:

Frontend: mostly Dash, with some ReactJS modules.

Backend: FastAPI , for authorization, file management and long calculation tasks. Basically, Dash is a Full-Stack Framework, doesn't need a Backend per se.

Python Worker with multiprocessing, with simple Redis Queue, for several long calculation tasks. "Long" used to be up to 1.5 hours in legacy code. managed to reduce it to <1min and the number of users never exceeded 5-6, so it was entirely reasonable to calculate everything inside Backend, but I've left the architecture as is for potential future developments.

MySQL + collection of Parquet files in S3.

Project was deployed on Client's infrastructure, Gitlab + OpenShift.

Education Stage

After taking several online Python courses, I've discovered Codewars, quickly reached 2kyu level, solving all kind of algorithmic katas. And inevitably fell victim of Dunning-Kruger effect 😑

Then I've decided to go for Machine Learning and took several online courses, Andrew Ng and such, culminating in self-study of CS231n. Didn't do any ML in real life yet, but still retain fond memories of coding a CNN in Numpy 😊

Client Projects, worth mentioning

Agent-based Financial Exchange Simulator, in Python
This is in fact an unsolved theoretical problem, as I had discovered while trying to build a robust non-degenerating model.

A market-making algorithm for a crypto exchange , which I accidentally coded in Scala. Had to learn the language on the fly.

Back-testing and parameter tuning of an Equity Trading Strategy, which started as a short script and developed into a small app with web interface running on AWS Beanstalk, mostly in Python with some CSS, picked along the way.