I build end-to-end real-time control and machine-learning systems across custom hardware and software: embedded firmware, wireless communication, computer vision, and reinforcement learning. Programming for 7+ years, professionally since 2021.

Recent projects: two from-scratch quadcopters with custom ESP32 flight controllers. KattVis uses onboard stereo vision for GPS-free motion and position estimation, while PidraQRL has a live RL agent tuning PID gains in real time on physical hardware.

🏆 Winner of the Yale SEA Most Outstanding STEM Exhibit - Unga Forskare 2026 National Final

A stereo-vision quadcopter built from scratch, flying without GPS or a barometer, using onboard stereo cameras and an IMU instead.

• Custom ESP32 flight controller with a cascaded PID loop at 250 Hz, running independently of the vision workload
• Deadcat airframe with a tuned front/rear roll-mix to compensate for the front motors' extra torque leverage
• Raspberry Pi 5 running an onboard stereo-vision and visual-odometry pipeline (dual OV9281 global-shutter cameras, 12 cm baseline) for GPS-free motion and altitude estimation
• Custom Flutter ground station with the actual CAD-modeled airframe rendered live, PID tuning, a per-motor test mode, and built-in stereo calibration
• Prototyped the stereo pipeline in Unity before buying the physical cameras, to prove the approach out first
• Custom nylon airframe designed in Fusion 360, manufactured and sponsored by PCBWay
• 10 seconds of controlled flight demonstrated so far; altitude hold functional and being tuned

Repo: KattVis

A quadrotor built entirely from scratch with a live SAC reinforcement learning agent tuning roll-controller gains in real time and a Unity HDRP digital twin mirroring live telemetry.

• Custom ESP32 flight controller with cascaded PID, Madgwick AHRS, and biquad filters (250 Hz loop)
• Full wireless chain: Flutter app -> BLE -> Raspberry Pi -> LoRa -> ESP32 (+ direct BLE for RL)
• SAC RL agent tuning PID gains in real time on the physical drone, no sim-to-real transfer
• Custom single-axis test rig for safe RL training with live propellers
• Unity HDRP digital twin with Fusion 360 models, mirroring live IMU telemetry over UDP
• Winner of the Yale SEA Most Outstanding STEM Exhibit - Unga Forskare 2026 National Final (Sweden's Young Researchers national championship)

🏅 Official winners list (Swedish) (Drone Control System)

🗞️ Winners announcement with photos (LinkedIn) (Swedish)

🌐 Digital exhibition

"The project stands out through a very high level of technical ambition. The custom-built flight controller, the distributed communication chain, and the well-designed failsafe solution demonstrate a deep understanding of real-time systems, control theory, and system safety. The methodology is exemplarily clear and reproducible." - Translated research jury feedback, Unga Forskare 2026

"In this project, the team members refused to take any shortcuts whatsoever. By rejecting ready-made frameworks and instead building everything from scratch, from communication to software, this work has demonstrated a technical dedication beyond the ordinary." - Winners catalogue, Unga Forskare 2026 (translated) - Link (Swedish)

📰 Press: Hulebäcksgymnasiet news (Swedish), School video (Swedish), Local press (Swedish)

Repo: PidraQRL

Generates a playable Geometry Dash level from any song using beat detection and physics simulation.

• Analyzes audio to detect beats and uses them to drive level generation
• Cube mode simulates ballistic arc physics reverse-engineered from observed in-game behavior
• Wave mode generates a diagonal path with rails and ramps sampled by arc length
• Reverse-engineered the .gmd format used by the GDShare mod to write valid level files importable directly into GD
• PyQt GUI with synchronized audio waveform and level geometry preview

Repo: GDForge

Software Developer - CPAC Systems - Recurring, 2021-2025

Five paid internships starting at age 15 across four years building internal tools and simulators in C#, Python, and Unity.

• Optimized a Unity simulator, doubling frame rate
• Built a Python diagnostics tool adopted by senior engineers for root cause analysis
• Built a license management service (HTML, CSS, JS)
• Developed machine simulations with custom mesh generation in Unity
• Refactored Perl codebases and worked on a web application (HTML, SCSS, TypeScript)

Embedded systems: C++, Arduino, PID control, IMU data filtering and handling, Fusion 360

Simulation & Graphics: Unity, C#, procedural generation (noise combining), Blender

Computer Vision: OpenCV, stereo vision, visual odometry, feature tracking, camera calibration

App: Flutter (dart), React, Python

AI: Reinforcement learning, YOLO, Torch, dataset generation

Languages: Swedish (native), English (fluent), French (basic)