top of page
Dec 2, 2025, 10_36_53 AM.png

PORTFOLIO

While we met at Yaak in automotive, AI and Machine Learning, we have our origins in game dev - but our range spans much further.

​

Find below a selection of projects we have worked on.

​​​​​

​

Vehicle Data Collection and Processing Kit

  • ​Overview

  • ​Hardware Architecture

  • Software Architecture

  • Backend: Event and Data Ingestion Platform 

  • 360° Video Processing Pipeline

  • Internal and External Drive Visualization

​

Next Level Procedurally Generated VR Simulation

  • Overview

  • Architecture

​​​​

Games (Selection)

  • The Cycle: Frontier​

  • Dead Island 2​

  • Dreadnought ​

  • Exoborne​

  • Grand Theft Auto: Chinatown Wars​

  • Grand Theft Auto IV​

  • June’s Journey​

  • Spec Ops: The Line​

  • Wolfenstein II: The New Colossus

​

Other Projects

  • Education

  • Health

  • Defense

  • Computer Vision and Digital Assets

  • Mobile OS

  • iGaming​

Projects

VEHICLE DATA COLLECTION AND PROCESSING KIT

  • Low cost central compute and sensor kit (8 cameras, CAN, IMU, GNNS) - easily installed in less than a day by any car workshop

  • OTA software updates and remote debugging

  • Mobile app for recording control and scene characterization

  • Producing and uploading terabytes of daily data for ML model training

  • Deployed on 60 cars in Germany, driving all day every day in any condition

HARDWARE ARCHITECTURE

  • Custom PCB design for carrier board

    • ESP32

    • IoT 4G modem

    • Access to ignition and CAN bus

  • NVidia Jetson Xavier carrier board on top

    • GSML extension board for 8 cameras

    • CAN bus access

    • GNSS receiver

    • 5G modem

    • SSD

  • Ethernet

  • Multiple hardware iterations

SOFTWARE ARCHITECTURE

  • Self developed software components mainly written in Rust

  • Software stack could run standalone: telemetry replay possible, interacting with backend

  • ESP32 firmware

    • Communicates with Jetson via uart

    • Monitor CAN bus and ignition

    • Communicate to IoT modem via serial line

  • Linux on Jetson

    • Main application

      • Video recording via gstreamer

      • Read car telemetry form CAN bus

      • Handle backend-initiated recording sessions

  • 3rd party services: OTA, logs, telemetry

BACKEND ARCHITECTURE:
EVENT AND DATA INGESTION

  • Backend

    • API in Go utilizing Postgres wrapped in Docker
      Running on AWS Fargate but could be deployed anywhere

  • Networking

    • Vehicle connectivity via internal network relying on Zerotier

  • Cloud/Infrastructure

    • Data upload from vehicles to S3

    • Data job orchestration via AWS Batch

    • Processing models self-contained on docker running provided from the ML team

    • Vehicle-kit software stack for E2E tests

    • Completely IaC-driven with Terraform

360° VIDEO PROCESSING PIPELINE

  • Model-assisted generation of calibration files

  • Machine learning assisted video processing pipeline in the cloud running on demand

  • Uploading video to S3 via API, ingesting metadata

  • Testing for video file completeness; triggering processing:

    • PII removal

    • Projecting camera images onto a sphere, which is aligned with the help of the calibration files

    • Compressing video further and making it streamable to websites and mobile devices (m3u/ts)

INTERNAL AND EXTERNAL DRIVE VISUALIZATION

  • Initial internal tooling

    • Individual selection of camera views

    • Telemetry for speed, indicator, headlights, pedal press intensity, instructor pedals

    • Drive event evaluation

  • External Tool: App

    • Ride overview & events, tracked in real-time via GPS

    • Automated 360° video creation with telemetry overlay

    • Statistics & progress tracking

NEXT LEVEL PROCEDURALLY GENERATED VR SIMULATION

  • Procedurally generated from real-world data, 3D rendered, and fully customizable

  • Creates digital twins of areas within a week

  • Full control over dynamic environment populated with AI-driven actors (e.g. traffic systems, pedestrians, events) and bespoke training scenarios

  • Broad range of use from special-purpose vehicles, military, autonomous driving or public driving education

VR SIMULATION: ARCHITECTURE

  • Based on Unreal Engine

  • Import OpenStreetMap and Mapillary data

    • Roads, lanes, intersections, yields, houses, …

    • Transforms data into generic format

  • Uses Houdini to generate levels

    • Data imported from OpenStreetMap as ground truth

    • Takes additional info about types of areas

    • Aims to be usable for driving schools

  • Tools to optimize fixing inconsistencies in input data

    • Automation used when possible

    • Convenient tools for level designers to determine issues and fix manually

    • Create workflows to prevent auto generation from overwriting changes done by humans

GAMES (SELECTION)

  • The Cycle: Frontier

​

  • Dead Island 2

​

  • Dreadnought 

​

  • Exoborne

​

  • Grand Theft Auto: Chinatown Wars

​

  • Grand Theft Auto IV

​​​​

  • June’s Journey

​

  • Spec Ops: The Line

​

  • Wolfenstein II: The New Colossus

OTHER PROJECTS

Note: due to effective NDAs we are not able to disclose a more detailed level of information on some of the projects listed below

  • EDUCATION: IARC - Global age classification tool for digital games and apps (licensed by Amazon, Epic Games, Google Play, Meta, Microsoft, Nintendo, and Sony PlayStation)

​

  • HEALTH: Game for hallucination research in psychosis patients

​

  • DEFENSE: HIL/SIL Simulation for autonomous drone development

​​​​

  • COMPUTER VISION & DIGITAL ASSETS: EyeEm - Computervision-driven global photography community and marketplace

​

  • TECH: Meltemi - Linux-based mobile phone OS (Nokia)

​

  • iGAMING: VEGAZ - Game engine for slot machines​

bottom of page