Justus Justus

Hi, I'm Justus.

I'm drawn to problems that require more than just code, systems that need to be reliable, data that needs to be trustworthy, and applications that feel effortless to use.


I build across backend, data, and native platforms, with a focus on thoughtful engineering and technology that solves meaningful problems.

Outside of software, you'll often find me cycling, hiking, or exploring ideas around renewable energy and sustainable mobility.

Experience

Timeline

Work & education

  1. ilium · Backend Developer / Data Engineer

    2025 – now

    Developing backend services and data pipelines, from data ingestion and processing, with a focus on reliability, data quality and maintainable architecture. This year, we displayed live vote counts for five state elections.

  2. Self-employed · Freelance Software Developer

    2023 – now
    • Design and develop custom web and mobile applications for small businesses.
    • Built RESTful APIs, authentication systems and automated data pipelines (Python, JavaScript), deployed with Docker and CI/CD pipelines on Firebase and Appwrite.
    • Improved customer operations through automation and tailored software solutions.

  3. HS Flensburg, Germany · B.Sc. Computer Science (dropped out)

    2022 – 2026

    Technical fundamentals in algorithms & data structures, software engineering, databases, computer networks, operating systems and mathematics.

Portfolio

Featured Projects

Applications, architecture & systems

Historical Election OCR

Data Engineering

Digitising historical election results from scanned statistical yearbooks.

Problem

German municipal election results from 1949–1969 mostly exist only as scanned typewritten or line-printed tables. Reading them automatically risks silent errors, and a plausible but wrong number is worse than a missing one.

Solution

A pipeline that accepts a value only when it is backed by evidence:

  • Table structure from two independent rule-based signals (ruling lines and column headers), so no column can shift silently
  • Multiple OCR engines read every number; a value is accepted only on full agreement, otherwise it goes to a human
  • Arithmetic checks per row (party votes sum to valid votes, etc.)
  • Full provenance: every value links to its image crop and every engine reading
  • No LLM in data extraction, so every value stays traceable

Results

Processed
7 volumes, ~1,000 table pages, 564,000 numeric cells
Fully machine-verified
8,891 result rows (everything else goes to a review queue)
Checked against official data (Hessen 1953)
7,094 of 7,095 values identical; the single difference is an error in the official file
Runtime
2.1× faster with parallel processing, output proven identical to the sequential run

Engineering

Typed data model with provenance for every value · domain knowledge as configuration (YAML), not code · regression tests on ~150,000 stored records · 131 tests, CI, Docker · documented architecture decisions.

Status

A first, self-contained part of a larger effort to digitise historical election data. It is being extended continuously with more states and elections and a self-trained recognition model.

Python Tesseract EasyOCR OpenCV PyTorch SQLAlchemy pydantic GitHub Actions Docker
Source Code

Projects

iPhone · iPad · Mac

A native task manager with Kanban board, list, table, calendar and timeline views per project, plus task dependencies, recurring tasks and custom fields. It syncs via iCloud and supports widgets, Siri/Shortcuts, Spotlight and actionable reminders. Fully localized in English and German.

What it demonstrates

  • Architecture: Domain logic, data layer and design system live in a separate Swift package, decoupled from the UI. Decisions are documented as ADRs.
  • Performance: Filtering and search run inside SQLite and stay fast with 10,000 tasks.
  • Product thinking: Accessibility (VoiceOver, Dynamic Type), onboarding, backup/restore, CSV export.
Swift 6 SwiftUI SwiftData CloudKit SQLite
Source Code

PDF Studio

Web & Desktop App

A PDF editor for the browser and desktop. Files are processed locally. Annotate, reorder pages, search and analyze documents, with optional AI summaries via local or cloud LLMs.

How & What was used

Built with React, TypeScript and Rust (Tauri), tested with Vitest and Playwright, deployed via GitHub Actions.

React TypeScript Rust Tauri Vitest Playwright GitHub Actions
Source Code

Landesparlament Monitor

Data Pipeline & Dashboard

Roll-call votes from a German state parliament, extracted automatically from official PDF transcripts, checked against the totals printed in them, and published as a searchable dashboard with an open API.

Problem

Individual votes in German state parliaments are buried in two-column PDF annexes, so you can't really search them.

Solution

A daily pipeline crawls the transcripts and extracts every roll-call vote with each member's ballot. The results go to a web dashboard and a REST API.

Trust

Each vote is checked against the printed yes/no/abstain totals. Anything that doesn't match is held back for human review instead of being published, and a hand-checked gold standard measures accuracy. Result: 33 of 33 votes match exactly.

Python FastAPI PostgreSQL Celery Redis Next.js 15 React 19 TypeScript Docker GitHub Actions

EcoAnalytics

Data & ML System

Machine learning pipeline and dashboard exploring renewable energy generation forecasts and sustainable mobility patterns.

How & What was used

Built predictive time-series models with PyTorch and Pandas for statistical data cleaning and feature engineering. Implemented API serving through lightweight Python microservices.

Python PyTorch Pandas Flask Data Science
Source Code

KitchenBook

Full-Stack Web App

A collaborative culinary web application to create, organize, and discover recipes with dynamic portion calculators.

How & What was used

Developed backend REST services in Java with Spring Boot and relational database modeling in PostgreSQL. Containerized the deployment with Docker and built a responsive UI with Vue.js.

Java Spring Boot Docker Vue.js PostgreSQL REST API
Source Code

Cardify

iOS Application

A modern flashcard application with a minimalist design, empowering students to retain knowledge efficiently through spaced repetition.

How & What was used

Built natively for iOS using Swift and SwiftUI following MVVM architecture. Implemented smooth interactive card-swipe gestures, offline-first data persistence with CoreData/SwiftData, and custom haptic feedback for delightful learning sessions.

Swift SwiftUI MVVM CoreData iOS
Source Code

Expertise

Skills & Technologies

Tools and stack I work with

Languages

JavaScript TypeScript Python Java Swift C#

Frontend & Mobile

SwiftUI SwiftData / CloudKit Tauri Vue.js Next.js React Tailwind CSS HTML5 / CSS Vite

Backend & Databases

Spring Boot Node.js Django Flask FastAPI REST APIs PostgreSQL MySQL Redis Celery SQLAlchemy

AI, Tools & DevOps

PyTorch MLX Core ML Pandas Matplotlib Docker Kubernetes CI/CD Nginx Google Cloud AWS Git / GitHub Figma JetBrains IDEs Claude Code Codex

Connect

Let's Work Together

Open for projects & conversation

Feel free to reach out directly for collaborations, questions, or just a quick chat about tech, sustainability, or software design.

GitHub

justus-dev02

LinkedIn

justus-dev

Discord

justus.dev

Email

justus-dev@proton.me