Amer Zavlan

• Designed and implemented the back end and the full AI processing solution for document-driven customs declaration workflows.
• Built multiple PDF processing modes to parallelize extraction paths and reduce end-to-end processing time for different document shapes and quality levels.
• Designed a graph-based execution model where each processing step writes results into graph nodes, supports fan-out and fan-in execution, and isolates process lifetime per node for better orchestration, recovery, and observability.
• Implemented automatic correction and recovery stages such as image deskewing, structural cleanup, and column re-classification when source documents were noisy or ambiguously formatted.
• Combined Laravel and Python into a single production workflow, using Laravel for application and back-end orchestration and Python for intensive AI/document-processing tasks.
• Built the system so that partial outputs could be refined by downstream nodes instead of forcing a single monolithic extraction pass.

• A Python-based lane detection project for self-driving car experimentation. I implemented the core computer vision pipeline using OpenCV to detect road lanes from static images, video input, and live camera feeds.
• A lightweight autonomous driving prototype focused on real-time visual processing. I developed the image analysis modules, integrated camera and video handling, and structured the project for testing across multiple input sources.
• An OpenCV-powered lane detector built as a practical self-driving car demo. I was responsible for the detection logic, preprocessing workflow, and end-to-end integration of image, video, and camera-based inference.
• A computer vision application for identifying lane markings in driving environments. I handled the implementation of the detection pipeline and connected the supporting components needed for processing and visualization.

This is an Angular and Ruby on Rails application that allows recruiters to better understand their employing pipeline. The app includes charts, data analysis, and several tools for management, executives, etc.

I was both a front-end and back-end developer. The project has an excellent codebase (unit tests, strongly typing where possible, etc.).

Openreel is an Angular and Node.js app that allows people to make movies on a webcam (usually interview-based documentary movies). The app contains all the necessary tools for people to record, stream, and edit videos.

I mainly worked on the front end and a little bit of the back end. Most tasks were related to WebRTC, video streaming, real-time data, etc. On the front end, it was an Angular app that used Angular Material, while on the back end, it used Nest.js

• A FastAPI-based web app for submitting text prompts and generating images through ComfyUI. I built the back end and front end using Python, Jinja2 templates, and vanilla JavaScript, including asynchronous job submission, polling, and image result rendering.
• A lightweight responsive interface for AI image generation with selectable output dimensions. I implemented the API flow, integrated ComfyUI workflow execution, and added a polling-based status system to keep the UI responsive while jobs run.
• A capable image generation tool with environment-based behavior switching. I added a demo mode that bypasses ComfyUI and serves a fixed sample image for previews, testing, and stakeholder walkthroughs.

The project is a Snowball Game in the Godot game engine. You roll downhill and avoid obstacles. A straightforward game but very entertaining. The map was infinite; therefore, all items were generated on the fly. Also, the game re-spawned every item now and then because we didn't want to run into floating-point precision problems after a long play.

The project, unfortunately, was halted mid-way and never released.

An Angular-based competition organization for mathematics. This is part of the global mathematics competition association: Association Kangourou sans Frontières, www.aksf.org.

I was in charge of developing the platform and some of the logistics as well, such as preparing tests for posts, etc. The front end is done on Angular, the design is a mix of Angular Material and Bootstrap, and the back end is done on Firebase, part of Google Cloud Platform.

There were a lot of scripts behind the scenes as well, most of them done in Python. They included features like automatically importing student data into the testing platform, parsing XLSX results, announcing news and results, generating diplomas, etc.

This is a 3D app related to architecture. They are making a data-driven approach to building management.

My part was to build a 3D app in canvas using Three.js to take 3D data and render it on screen. It was a proprietary model format. They also needed to stream the data from the server to the viewer more efficiently (for example, show only data relevant to the user's current camera location).

I built the project in TypeScript and transpiled it to JavaScript for the web.

An Angular-based competition organization for informatics, part of the Global Informatics Competition Association (International Challenge on Informatics and Computational Thinking: www.bebras.org).

I was in charge of developing the platform and some of the logistics as well, such as preparing tests for posts, etc. The front end is done on Angular, the design is a mix of Angular Material and Bootstrap, and the back end is done on Firebase, part of Google Cloud Platform).

There were a lot of scripts behind the scenes as well, most of them done in Python. They included features like automatically importing student data into testing platform, parsing XLSX results, announcing news and results, generating diplomas, etc.

An interactive water demo using the Three.JS library.
Approximation to water physics by simulating each pixel as a particle that has momentum, vertical position, and force from neighboring particles.

The height of water is then converted to a normal map and applied to a nice reflective plane.

A particle demo with trails using the Three.JS library. The gravity for each dot is calculated using Newton's gravity formula. Its position is then collected to display a line trail. This demonstrates the usage of manual mesh creation and management of its vertices.

A portal written in React and Python used for programming competitions. The project is backendless but heavily relies on many Python scripts (e.g., to set up a competition, start the virtual machine and import contestants). The database and login are provided by a Supabase project. Supabase is an open-source alternative to Firebase.

There are many useful front-end features, such as recording the screen during competition, a dynamic application form for a competition, profile management, OAuth login, charts showing scores over time, and more.

A Django, Angular, and Android system including multiple subsystems, such as printing tickets, real-time monitoring of entrances and employees, remote app updates, offline mode, data synchronization, user management, analytics, graphical analytics, and reporting.

A React-based form-builder app that's similar to Typeform. I was a front-end developer tasked to build the front-end from scratch. I utilized caching mechanisms and was careful to have a single source of truth by utilizing react-query where possible. Otherwise, it takes data from Redux. Modified Shadcn components and Tailwind CSS were used to improve the app's look and feel. TypeScript ensured that all of the data was strongly typed and conformed to back-end data.

This is a part of my Space Maze game. The entire game is available on my GitHub page.

This file demonstrates how I created a texture-based collision. The ball is controlled in a "vacuum" and has to go through a maze. The maze is constructed into textures where the transparent parts are insubstantial, and the opaque parts can collide.

The file focuses on optimizing the collisions and dividing them into Quadtree structures in order to optimize the search of opaque pixels and speed up the detection of a ball-pixel collision.

Note: In the file, I named the structure OCTREE instead of QUADTREE—this is yet to be changed.

How to run:
1. Obtain code from https://github.com/charlieamer/MazeMonoGame.
2. Obtain Xamarin IDE.
3. Open the project solution (.SLN file).
4. Download the latest MonoGame library from monogame.net/2017/03/01/monogame-3-6/.
5. Add monogame to the dependencies.
6. Optional. You can visually see the performance of Quadtree if you enable the DEBUG_VISUALISATION macro.