// Projects & Experiments · AI · Simulation · Tools

[Tomasz Wietrzykowski]
- Builder.

I build AI systems, local-first tools, simulations, and interactive experiments - from production-ready agents to generative art and cognitive models.

34+
Projects
10+
GitHub repos
100%
Open source
0
Cloud / subscription
Profile photo
⚙️
⭐ Flagship · Production
Enterprise RPA & Data Management Suite

A comprehensive Robotic Process Automation and Data Management ecosystem architected and built from scratch - acting as a central integration hub seamlessly bridging modern web portals, closed legacy desktop applications, local databases, and cloud services (Google Workspace).

Eliminates thousands of hours of manual labor by fully automating mass report downloading, unstructured PDF data extraction, spreadsheet synchronization, and dynamic document generation. Scales to 1.3M+ records without UI freezing.

📊 High-Performance Analytical Database
DuckDB + Pandas ETL pipeline. Ingests and queries datasets exceeding 1.3M records. Vectorized mass inserts, fuzzy search, and dynamic value mapping for instant retrieval despite typos or inconsistent formatting.
🌐 Advanced Web Automation
Async Playwright bots navigating complex React/MUI SPAs. Bypasses dynamically rendered DOMs, hidden iframes, and non-standard components. Bulletproof mass downloads via active polling and race-condition mitigation.
🖥️ Legacy Desktop Automation
Pywinauto + PyAutoGUI orchestrating closed-source legacy apps (Delphi/Win32) with no API access. Hybrid approach: native Windows API hooks combined with OpenCV template matching and OCR - resolution and DPI-agnostic.
📄 Intelligent Document Processing
PyMuPDF parses unstructured ticket/report PDFs. NLP + Regex auto-categorizes requests and populates the UI. Dynamically maps filtered DB records into .docx templates - generating print-ready legal protocols on the fly.
⚡ Mass File Search Engine
Thread-pool concurrent engine scanning gigabytes of raw text files in seconds. Instantly extracts matching records and exports them into cleanly formatted Excel spreadsheets via openpyxl.
🔐 Security & Background Daemon
AES encryption module for credential management across multiple platforms. Master password decrypts at startup and injects credentials into isolated child processes via os.environ. Custom daemon boots with Windows, evaluates schedules, and maintains detailed audit logs.
🤖
#02

Office AI Studio

Open-source, 100% local AI automation suite for office work. Smart Drop Zone for drag-and-drop actions, Pipeline Builder for multi-step workflows, Python script automation, Auto Tasks triggered by file rules, CSV/data tools, and Meeting Notes → tasks + email drafts. Works with Llama, Mistral, Bielik and any Ollama model.

Python Ollama Desktop GUI Office Automation
#33

GroqAgent - Natural Language Computer Control

Control your Windows 11 computer with natural language. Combines Groq's LLMs with a full tool suite: browser automation (click, fill forms, JS execution), file operations, Excel creation with charts, CMD/PowerShell, and web fetching. Features dual-model routing - auto-selects between fast (Llama 3.1-8b) and smart (Llama 3.3-70b) based on task complexity.

Python Groq API Windows 11 Browser Automation
#34

Agentflow - Autonomous AI Market Research Agency

A fully functional multi-agent system built with CrewAI and Google Gemini 2.5 Flash. Agents have live web access via DuckDuckGo search and collaborate in a hierarchical process: Manager → Internet Researcher → Market Analyst → Report Director. Outputs clean, professional Markdown business reports. 100% free to run using Gemini's free tier.

Python CrewAI Gemini 2.5 Multi-agent
🧠
#06

Bio-Brain - Biologically Faithful Artificial Brain

A spiking neural network with real Izhikevich neuron dynamics running a virtual creature that learns to survive. 29 spiking neurons, STDP learning (from Bi & Poo 1998 rat hippocampus research), dopamine/serotonin reward system (three-factor learning rule), and Dale's law enforced (excitatory/inhibitory segregation). No backpropagation - purely local spike timing and neuromodulation. One Python file.

Python pygame numpy Spiking NN Neuroscience STDP
#08

H-Map - Cognitive Memory System Inspired by the Hippocampus

An experimental cognitive app that builds a dynamic memory map from live camera input using 400-dim vector embeddings (digital engrams). Visually similar frames cluster into artificial Place Cells, forming a navigable cognitive map. Includes a predictive layer that anticipates upcoming visual states and a reinforcement layer (reward/punishment) that shapes system behavior. Real-time, self-organizing - no neural network classifier.

JavaScript Vector Embeddings Place Cells Cognitive Science
#20

Biologically Inspired Mouse Navigation Simulator

Biologically inspired navigation agent using place cells, grid cells, head-direction cells, and SLAM-like memory. The agent explores an arena, avoids obstacles, builds a spatial map, and finds food autonomously - all driven by the same cell types found in mammalian hippocampus.

Python SLAM Place Cells Grid Cells Neuroscience
#11

Consensus AI - Dual-Agent Deliberative Reasoning

Instead of instant answers, two AI personas start from opposite viewpoints on your problem, debate, challenge each other's arguments, refine positions, and converge toward common ground. Useful for difficult decisions, complex tradeoffs, and exploring multiple perspectives before reaching a conclusion.

JavaScript Multi-agent Deliberation
📡
#12

Collision Field AR - Real-Time Obstacle Detection

AR app that perceives surroundings in real time using a single monocular camera in a mobile browser. Detects walls, floors, furniture, stairs, and drop-offs; builds a local 3D occupancy grid with an attention mechanism that updates only currently observed regions. Provides visual red-edge warnings, haptic feedback, a live top-down mini-map, and records 3D camera trajectories. No LiDAR, no cloud, no app install.

JavaScript WebGL Computer Vision SLAM-like Mobile AR
#03

EcoSound AI - Ecosystem Health Analysis Through Sound

Assess ecosystem condition from a .wav file or live microphone. Computes scientific acoustic indices (ACI, ADI, BIO, NDSI), scores ecosystem health 0–100, generates a dashboard with spectrogram/waveform/charts, optionally identifies bird species via BirdNET (Cornell Lab), and compares multiple recording locations side by side. Pure Python, runs locally.

Python Bioacoustics BirdNET Ecology
#16

Mix & Mesh - Single Photo to Interactive 3D Scene

Upload one photo (room, landscape, portrait, object) and the AI instantly generates a depth map, creates semantic segmentation masks (recognizing objects), and fuses them into an interactive 3D scene. Rotate the view, look from different angles, apply effects, or export to 3D format.

JavaScript Depth Estimation Semantic Segmentation 3D Visualization
#17

RedWall 3D Architect - Room Photo to 3D Floor Plan

Upload any room photo (even low-quality phone shots) and AI reconstructs the 3D structure, estimates camera position, and recognizes furniture with approximate dimensions. Navigate the scene in 3D, switch between realistic and semantic views. Powered by Gemini 3 Flash. Useful for interior designers and real estate agents.

JavaScript Gemini Vision 3D Reconstruction Interior Design
#13

Obstacle Map - Real-World GPS Traversability Map

A real-world GPS map that models traversable space, not just routes. Features live GPS tracking, 3D building extrusions from OpenStreetMap, and a collision/occupancy heatmap. Built for spatial cognition research, robotics, AR, and accessibility applications.

JavaScript WebGL OpenStreetMap GPS Robotics
#18

Drone Cloud - Aerial AI Spatial Scanning

Using a drone and AI, create spatial scans of real environments analyzable for distances, spatial orientation, thematic segmentation of objects/sizes, and occupancy grids using Python. Demonstrates depth colorization, semantic segmentation overlays, and 3D point cloud generation from drone footage.

Python Drone Point Cloud Spatial Analysis
#32

Robot SLAM Simulator 3D

Interactive SLAM demonstration where clicking any object in a 3D room causes a robot to automatically plan a path, avoid obstacles, and update its environmental map in real time. Shows SLAM algorithms in practice - from robot localization to A* path planning.

JavaScript Three.js SLAM Pathfinding
🧬
#15

Bio-Evo Sim - Interactive Bio-Evolutionary Particle Simulation

Particles governed by basic interaction rules (attraction/repulsion by type) spontaneously form complex self-organizing structures resembling living organisms - from simple clusters to ecosystems with predation, symbiosis, and reproduction. Adjustable parameters (forces, velocities, friction, particle types) enable exploring infinite "worlds" and discovering new digital life forms.

JavaScript Emergence Artificial Life Evolution
#03 (alt)

3D Biomorphs - Dawkins' Blind Watchmaker in 3D

A recreation of Dawkins' classic Biomorphs from The Blind Watchmaker, now with behavior mutations in 3D. Simple genetic parameters evolve shapes and movements illustrating how cumulative selection produces complexity without foresight. Also includes the original 2D Biomorphs version as described by Dawkins.

JavaScript Genetic Algorithms Evolution 3D
#14

Evolution 3D & PROTOS - Emergent Life Simulations

Two particle-life simulations: Evolution 3D - 3D particles with interaction rule matrices that spontaneously form molecule- and orbital-like patterns; press * for new random rules. PROTOS - 2D emergent life where simple particle interactions produce creature-like behavior. No scripts, no agents - pure bottom-up emergence.

JavaScript Particle Life 3D Emergence
#22

MechEvo - Mechanical Creatures That Dock & Evolve

Mechanical creatures with physical joints and sensors that dock onto each other, form increasingly complex mega-structures, and evolve over generations. Emergent locomotion strategies arise from physical constraints and selection pressure.

JavaScript Physics Evolution Procedural
#31

Virtual People Simulation - AI-Driven Social World

Observe the interactions, decisions, and lives of AI-driven people in a virtual world. Each character makes autonomous decisions about where to go and what to do, interacts with locations and other characters, and adapts to random weather events. Thought bubbles reveal their real-time activities, emotions, and social connections.

JavaScript Agent Simulation Social AI
#30

Text-to-Creature - Your Text Becomes Life

Each letter in your text is DNA: vowels = muscles, consonants = bones, numbers = sensors. Different text sequences create different 3D organisms that crawl, pulse, and move with realistic physics simulation. Type any string and watch it become a living creature.

JavaScript Physics Procedural Generation 3D
🎮
#09

Nexus-MUD - Infinite AI-Generated Text World

Each play-through generates a new universe - unique, unpredictable, governed by its own logic. Explore a vast chessboard-like map; every step generates a new location, landscape, or story in real time via AI. Every location, character, and event is permanently stored in world memory. The further you travel from the starting point, the stranger the world becomes - from classic fantasy to surreal abstract realities.

JavaScript AI Persistent World MUD
#25

Exoplanet Explorer - Procedural Alien World Discovery

Explore a procedurally generated universe of exoplanets filled with strange lifeforms and unexplained phenomena. Discover unique ecosystems, encounter rare anomalies, and capture extraordinary creatures - each with detailed characteristics. Scan, collect, and build your own cosmic archive as you travel across distant star systems.

JavaScript Three.js Procedural Exploration
#19

Interstellar - Universal Atlas 3D

Interactive 3D atlas of the universe with seamless scale transitions - from atomic and molecular structures through cells, grass, planets (Mercury to Neptune), star systems, the Milky Way, and large-scale cosmic structure. Spatial exploration with a unified model of the cosmos, real-time audio systems, and clickable object info panels.

JavaScript Three.js 3D Science
#26

Shake It! Cataclysm - Physics Destruction Simulator

Choose your force - earthquake, tornado, or meteor - and watch the city face it in real time. Shake your smartphone to trigger earthquakes. Dynamic physics, collapsing structures, shockwaves, and debris make every scenario unpredictable. City Integrity meter tracks total destruction progress.

JavaScript Three.js Physics Mobile
#21

Flight Simulator - Fly Anywhere on Earth

A realistic browser-based flight simulator with real-time dynamic weather, a seamless day-night cycle, and hundreds of detailed airports from major hubs to mountain strips and island runways. Complete freedom to choose route, aircraft, and conditions.

JavaScript Three.js Flight Sim Browser
#29

DJI Drone Simulator - Realistic Browser Drone Flight

Fly a DJI drone in a large realistic 3D world with buildings, trees, and terrain. Physically accurate flight model, realistic FPV and third-person camera modes, Unreal Engine–level visuals - entirely in the browser. No installation.

JavaScript Three.js Physics FPV
#15b

AI Kitchen Sim - Webcam-Controlled VR Kitchen

Step into a realistic kitchen in first-person view without controllers or VR headsets. Advanced head and hand tracking via a standard webcam lets you look around, control virtual hands, and physically interact with objects - grab pots, open cabinets, stir ingredients - with realistic physics.

JavaScript MediaPipe WebXR Physics
#27

Fur Ball - Tactile Relaxation Experience

A calming, tactile browser experience. Gently pull, brush, and stroke a soft furry sphere rendered with real-time strand simulation. Multiple modes including grass and magma. Designed as a simple, soothing interaction for stress relief and focus.

JavaScript WebGL Interactive Art
#24

Quantum Orbitals 3D - Real Wave Function Visualization

Interactive 3D visualization of atomic orbitals showing electron probability distributions around the nucleus. Each particle's position is calculated using real quantum wave functions from Schrödinger's equation. Rotate and explore different orbital shapes by adjusting quantum numbers (n, ℓ, m).

JavaScript Three.js Quantum Mechanics Science Viz
📚
#05

edu-transformer - GPT Architecture in One Python Script

A free, hands-on GPT course in a single Python script. Same architecture as GPT-2/3/4, just 8K parameters instead of 1.8T. Run it instead of watching 4-hour lectures: build/train/generate text in 3 seconds, debug mode showing every tensor and attention weight, compare tiny/small/medium models side-by-side, run ablation studies, take quizzes, and use an interactive playground. No GPU needed.

Python PyTorch Transformers GPT MIT License
#07

OneLook - Visual Answer Engine

A new kind of AI experience where answers are delivered not as text but as a single visual. Enter any question and instead of reading paragraphs, you instantly see an AI-generated image representing the idea behind your query. Fast, intuitive, entirely visual - turns questions into pictures.

JavaScript AI Image Generation UX Experiment
🛠️
👁️ EyeSpeaker

Text input using eye gaze - no keyboard, no mouse. AI predicts next words and phrases; select them just by looking at the screen. Designed for users with limited mobility.

✍️ AI Handwriting Notes

Write with your finger directly on the screen - AI instantly recognizes letters and sentences in real time and converts them to clean text. No keyboard, no stylus.

🗣️ SpeakNow

Real-time text-to-speech communication aid for people who cannot speak. Type a message and it is instantly spoken aloud - fast, simple, barrier-free.

📳 Haptic Nav

Screen-free haptic navigation for visually impaired users. Silent when on route - increasing vibrations when deviating. No screen needed, works entirely through touch feedback.

🎭 AI Character Animator & FaceMorph

Load any image - background is removed and the character is animated. FaceMorph mirrors your real-time facial movements onto any photo. Record and download in seconds.

🦾 Hand-Controlled Robot Arm

Control a virtual robotic arm by moving your hand in front of the camera. Works on both laptop and smartphone rear camera. A teleoperation prototype built entirely in the browser.

🏗️ AR Mesh Builder

Build a 3D model of your surroundings through your phone camera - tap to place points and walls, walk around to view from different angles, export as GLB. No ARCore or SLAM required.

📷 Depth 3D

Real-time depth map generation from your camera feed. Visualizes scene geometry live - useful for computer vision experiments and spatial understanding demos.

🚧 Obstacle Detector

Uses your phone camera and flashlight to detect obstacles in real time. Measures brightness differences between frames to estimate relative distance - no depth sensor needed.

🎹 Virtual Piano

Play a piano using only your hands in front of the camera - no touch required. Two versions with different interaction modes. A creative gesture-control music experiment.

🖱️ Windows Hand Gesture Simulation

A prototype for controlling a desktop interface using hand movements and camera - no mouse, no keyboard. Pinch gesture performs a click. A proof-of-concept for gesture-driven OS control.

🔬
#10

PersonaX - Customizable Real-Time Virtual Persona

Talk in real time with a virtual persona. Upload a face, set a personality, tweak the voice, and chat live. Supports 3 languages, fully customizable characters including face detection fine-tuning and voice selection.

JavaScript Face Tracking TTS AI Chat
#28

Advanced 3D Web Environment - Unreal-Level Browser World

A full Unreal Engine–style interactive 3D world running entirely in the browser. Swim, fire lasers, watch procedurally animated animals, and explore dynamic weather systems - all in real time, no installation.

JavaScript Three.js 3D World Procedural