Tomasz – Python, React, Django

AI agent performing tasks using a fully controlled browser. The user would define a task to perform using natural language (e.g. order a pizza to my home), the agent would spawn a browser on a server and do the thing.

• worked on browser integration: iframe handling (merging all frames into a single, consistent tree of nodes); simplifying the internal representation based on computed CSS (deciding which items can be merged together to save tokens without losing any information); helping the agent handle less standard websites (heuristics / deriving information from CSSOM); scraping performance improvements (optimizing the server <-> Chromium communication);
• implemented integration with Stripe;
• infrastructure setup, CI/CD integration;
• conducted e2e testing;
• was in charge of code quality and tooling;
• created a data annotation tool for human labellers (for a model finetuning pipeline)

A fleet monitoring, analytics, and optimization platform. It aims to digitalize the work of a fleet manager with features such as:

• real-time fleet monitoring (through 3rd party hardware providers);
• historical fleet performance analysis;
• finding optimal routes for a constrained fleet moving goods within a certain area (Vehicle Routing Problem).

• designed front-end and back-end architecture;
• designed and implemented all CI/CD and infrastructure pipelines;
• overlooked the entire engineering process;
• directly managed a team of developers, indirectly managed +-15;
• took part in marketing, UX, and business processes;
• designed integration with third-party providers;
• took an active part in day-to-day development activities, from implementation to code review and testing;
• mentored less experienced team members.

An NGO promoting transparency in governing bodies. It's a web app citizens can use to ask for legal advice related to the cause.

• introduced E2E tests using Cypress;
• implemented a prototype Business Intelligence module;
• implemented autocomplete for forms;
• fixed bugs, both front-end and back-end;
• optimized database queries.

Cloud Autoscaling is responsible for automatically increasing the size and count of (virtual) machines allocated to a given workload based on the current state (e.g. CPU usage). Predictive Autoscaling enables producing autoscaling recommendations ahead of time, based on historical usage, using proprietary machine learning tools.

• was the main contributor (100k+ LOC) of C++ services processing incoming metrics, gathering historical data, interacting with the machine learning library, and producing predictive recommendations in real-time;
• designed, discussed, and implemented changes to other teams' codebase, as required to ship our product;
• designed unit, integration, and e2e testing scenarios;
• integrated new microservices with existing monitoring and alerting systems (Google internal stack, but similar to products available in GCP);
• mentored 3 interns working in sibling teams.

A platform connecting teachers with parents. It allows users to create local social networks around classes and schools, making it easy to communicate, share events, and get to know each other.

• implemented real-time chat and notification features using Web Sockets, both on the frontend and the backend;
• extended email notification features;
• searched for and fixed database performance bottlenecks;
• improved tooling and test coverage on both ends.

Prime Video Living Room Playback is a team responsible for making the Prime Video app available on devices such as smart TVs, game consoles, set-top boxes, etc. The project he was working on aimed to unify the codebase for all devices, effectively implementing a custom web browser for all supported devices and shipping the app as a JavaScript bundle.

• implemented core Web APIs in JavaScript on top of our custom JS runtime;
• glued together rendering, engine, and client layers using a custom coroutine implementation;
• helper other team members transition from C++ to JavaScript;
• diagnosed a PoC implementation at the beginning of the project to check if the approach is viable.