Maksymilian Grela · engineering + software
I study Automation and Robotics alongside a second degree focused on Industry 5.0. I build software around real workflows, from CNC timing to study automation, and use small projects to connect code, mathematics and physical systems.
I like understanding things by getting my hands on them.
At Gdańsk University of Technology, I am studying Automation and Robotics (semester 3) alongside a second degree focused on Industry 5.0 (semester 1).
My interests sit somewhere between software, AI, automation and physical systems. I am deliberately keeping the map wide for now, not because I want to collect technologies, but because I like seeing what happens when different fields start clicking together.
I learn something, build something, break something, write it down, and then usually find another thing I want to build.
working tools, experiments and the next steps
A multi-machine CNC timer built around a real shop-floor workflow. A lightweight PWA tracks machining time, while a serverless notification pipeline schedules Web Push reminders.
The first working module of Micro Jarvis, also usable on its own. A terminal workflow that turns lecture materials and personal notes into structured Obsidian notes with Codex CLI. Built for my Windows/WSL setup, with configurable detail and a repeatable installation process.
Identifying a simple thermal model from noisy synthetic temperature and load data. I compared least-squares estimation with gradient descent, simulated the recovered dynamics and checked the trajectory error.
A Python pipeline for collecting articles, normalizing URLs, deduplicating, filtering and exporting Markdown. An asynchronous loading update measured roughly 5.5× faster than the earlier version in a project benchmark.
A Python CLI for organizing files, with an emphasis on separating filesystem operations from decision logic and testing edge cases.
A modular terminal work hub, designed as independent tools that can run on their own or plug into one control center. The first existing module is lecture-notes-codex; a Python/SQLite task-control module is the next planned piece.
A small real-world business built around handmade rosaries. Domain, Shopify, social media, operations and eventually as much automation as actually makes sense.
A tiny physical interface to an AI assistant. Press a button, talk to the phone, let the agent do something useful. The ring is the fun part; the more serious idea is the open interface underneath it.
A small additional lock controlled by NFC. Part security experiment, part embedded project, part excuse to learn how electronics and mechanics behave when you put them on an actual door.
A compact IR heating experiment with thermal feedback. The bit I really want to test is an autofocus loop: move the lens, measure the thermal spot, adjust, repeat.
A selective community for ambitious IT/AI builders. The actual idea is not another Discord server; it is a dense enough network of people that building together becomes easy.
the machinery behind the curiosity
20 Under 20 · personal learning programme
6/20 foundation milestones completed · continuing
Six completed stages of a broad technical base. I keep testing what I learn through small projects.
CLI tools, data pipelines and async I/O. Practical evidence: File Organizer CLI and News Digester Mini.
Branches, commits, pull requests and repository workflows across two machines.
Bash, shell workflows and a developer environment. Practical evidence: DotFiles Mini and study automation.
Foundations: schema design, primary and foreign keys, constraints, 1:N and N:N relationships, SQLite/Python basics, migrations and seed data. Advanced queries remain a next step.
Practical Big O; list, set, dict, deque and heap. Exercises include bracket validation, top-N sorting versus heapq, iterative tree traversal and BFS shortest paths; A* appears in my labs.
Numerical linear algebra, SVD and conditioning, gradients and gradient checking, Euler/RK4, stability and model identification. Practical evidence: the synthetic-data thermal model project.
I define the problem and constraints, delegate parts of the work, read the results and diffs, then test or benchmark and document what changed. News Digester and lecture-notes-codex are practical examples of this workflow.
After a starter covering days 001–014, I moved to selective monitoring of lasting changes in models, agents, protocols, coding tools, infrastructure and safety.
things currently occupying too much RAM
The interesting jump is not “smarter chat”. It is software getting enough agency to choose tools, observe results and keep going.
Sometimes the best optimization is changing what the system is trying to do rather than polishing the current solution.
Complexity has a cost. Removing a component can be a better engineering decision than making the component better.
A recurring thought from my notes: if everything you try works, your experiments might not be ambitious enough.
what is actually happening
Code when the problem is software. Electronics when software is not enough. AI when the workflow can be delegated. Engineering when physics has the final word.
October 2026: two engineering degrees, practical CNC work, study-workflow automation and mathematical modelling. I am refining lecture-notes-codex and exploring how independent tools can fit into Micro Jarvis.
grela.tech
There is no grand final version yet. I just keep finding the next thing worth understanding.