# Raphaël Mitas

> Full Stack Developer, Germany

Full-Stack Developer with experience at Mercedes-Benz and Audi, now driving innovation at Kimeta. I specialize in the React/Next.js ecosystem with TypeScript, tRPC, and React Native, building type-safe applications from frontend to backend.

What sets me apart: I don't just use AI tools, I build with agents. At Kimeta I introduced Cursor and Claude Code and built agentic automations that turn tickets into merge requests and review code automatically. Our entire Cypress e2e suite moved to Playwright in a single day, without compromising on code quality.

Outside of coding, you'll find me on the slopes skiing, out hiking, behind the camera capturing moments through photography, or at the table for a board game night.

- Email: mail@raphaelmitas.com
- GitHub: https://github.com/RaphaelMitas
- LinkedIn: https://www.linkedin.com/in/raphael-mitas-5144b510a/
- Location: Germany
- Download Resume: https://raphaelmitas.com/cv.pdf
- Languages: German (Native), French (Native), English (Fluent)

## Experience

### Full-Stack Developer, Kimeta

2024-06 – Present · full-time · Darmstadt, Germany

Full-stack development with a frontend focus for a multi-tenant job portal platform. Maintains the team's Next.js application and owns its CI/CD.

- Early and enthusiastic adopter of agentic coding: introduced Cursor and Claude Code and shared workflows hands on. The team adopted them with real success, with code quality staying high.
- Introduced e2e testing at the team and automated it in the CI pipeline, then rewrote the entire Cypress suite in Playwright in a day with agents. Validated the migration with fault injection and by running both suites in parallel in CI until the new one had proven itself, with adversarial review agents over the diff.
- Built and deployed an agentic coding platform used by the team: a Tauri app on the Claude Agent SDK and Codex SDK with custom MCP servers over YouTrack and GitLab, where agents take tickets, open merge requests and run code reviews, each in its own Docker environment.
- Developed an AI chatbot and other AI features for the platform.
- Co-led a Next.js (TypeScript) codebase overhaul with one other engineer, introducing atomic design for reusable components across the multi-tenant architecture, and has maintained the application since.
- Played a key role in preparing the migration of a legacy PHP system to the team's Next.js platform.
- Built automated Lighthouse performance monitoring for all company sites: scheduled data collection, an Express.js API, Grafana, and a Next.js dashboard with tRPC and shadcn/ui.
- Owned the team's CI/CD pipelines (GitLab CI, Docker Swarm), developing and enhancing them to automate build, test and deployment.
- Collaborated closely with the backend team to orchestrate the migration of multiple platforms to a new, unified architecture.
- Managed the Kimeta Expo App (React Native), a browser-wrapper mobile application with native feature integration.

Skills: TypeScript, React, Next.js, Tailwind CSS, tRPC, shadcn/ui, React Native, Expo, Express.js, Agentic Coding, Claude Code, Cursor, Playwright, Cypress, GitLab CI, Docker Swarm, Grafana, Problem Solving, Communication, Cross-functional Collaboration, Adaptability, Self-Learning, Mentoring

### Full-Stack Developer, Mercedes-Benz

2022-04 – 2023-06 · working-student · Sindelfingen, Germany

Developed real-time analysis and predictive maintenance solutions for autonomous guided vehicles (AGVs) in factory environments.

- Built and deployed a real-time analysis and error prediction application for over 400 autonomous guided vehicles (AGVs) using Next.js (TypeScript) within the factory environment. This initiative significantly elevated the monitoring and predictive maintenance capabilities.
- Transformed a legacy Plotly Dash application to Next.js with Material-UI, developing live production dashboards for factory operations.
- Implemented REST services using FastAPI (Python) to facilitate efficient data exchange and integration with the backend systems, ensuring seamless operation of the real-time analysis platform.
- Contributed to backend development, creating scalable and resilient services using Python, Redis, Kafka, and PostgreSQL. This underpinned the real-time data processing and error prediction functionality for the AGVs.
- Leveraged Kubernetes and Docker within a DevOps framework to deploy and manage both frontend and backend components, running on Azure cloud environment.

Skills: TypeScript, React, Next.js, Python, FastAPI, Redis, Kafka, Kubernetes, Docker, Azure, Problem Solving, Communication, Adaptability, Self-Learning, PostgreSQL, Material-UI, Plotly Dash

### Frontend Developer, Audi

2020-08 – 2021-02 · internship · Neckarsulm, Germany

Developed business intelligence dashboards and ETL solutions for workplace safety evaluation.

- Developed an Oracle Apex dashboard for workplace safety evaluation with advanced filtering and drill-down/drill-up capabilities.
- Designed and implemented user interface improvements for the Industrial Engineering dashboard, enhancing usability and visual appeal.
- Employed Knime as an ETL platform for periodic data transformation and integration, ensuring the Oracle database was updated with the latest business intelligence.
- Optimized Excel VBA workflows for automated data processing and reporting tasks.
- Mastered complex SQL queries and PL/SQL programming, a necessity given Oracle Apex's reliance on PL/SQL for nearly all database interactions and application logic.
- Enhanced business intelligence and operational efficiency by orchestrating the periodic transformation and seamless integration of data sources into a unified Oracle database framework.
- Integrated the Industrial Engineering application into VW's production environment in Hungary, enabling cross-site deployment.

Skills: Oracle Apex, Oracle, SQL, PL/SQL, Knime, Problem Solving, Communication, Adaptability, Self-Learning, Excel VBA

## Education

### Bachelor of Science, Information Systems (Wirtschaftsinformatik und Digitale Medien)

Stuttgart Media University · 2018-10 – 2023-10

Comprehensive program combining business informatics with digital media technologies. Bachelor thesis: "Use of Generative AI in Software Projects - Analysis of Effectiveness and Applicability"

### Bachelor studies, first year, Information Systems (Wirtschaftsinformatik)

Technical University of Munich · 2017-10 – 2018-09

First year of Information Systems, then switched to Stuttgart Media University for a more hands-on, project-driven program.

### Abitur, Data Processing & English

Heinrich-Emmanuel-Merck-Schule · 2014-10 – 2017-06

German secondary school diploma with focus on data processing and programming.

## Projects

### Photopipe

Free, open source macOS app for photo culling and editing. Denoised RAW previews, ratings, curves and crop in one native tool. Architected and shipped solo, with coding agents under my review.

- View Live: https://photopipe.net
- View Code: https://github.com/RaphaelMitas/photopipe
- Tech Stack: Swift, Tauri, Rust, React, TypeScript, Tailwind CSS, shadcn/ui, Playwright, Claude Code, Agentic Coding

Photopipe is a free, open source macOS app for photographers who come home with thousands of RAW photos and need to find the keepers fast. It covers the whole workflow in one native tool: cull with denoised previews the moment a folder opens, rate with the keyboard, edit with sliders, tone curves and crop, and export finished JPEGs. The disk stays the source of truth, with ratings and edits written into XMP alongside the files, so nothing is locked into a catalog or a cloud.

The app is a Tauri 2 shell around a Swift core that drives Apple's native RAW pipeline for decoding and denoising, with an AI rating feature built on Apple's aesthetics scoring to presort a shoot. I architected and shipped it solo, with coding agents doing implementation under my review, adversarial review agents on every merge and verification agents that must reproduce a finding before a fix lands. The native rebuild went from an empty repository to a signed, notarized v0.1.0 in 48 hours, followed by nine releases in the following eight days. The full story is on the blog.

- Culls a 3,000-photo shoot with denoised RAW previews via Apple's native RAW pipeline, with no batch step and no waiting
- Full edit pipeline: exposure sliders, per-channel tone curves, crop and straighten, persisted as XMP so files stay the source of truth
- Instinct AI rating presorts a shoot using Apple's aesthetics scoring
- Rebuilt on Tauri and Swift in 48 hours to a signed, notarized v0.1.0, then nine more releases over the following eight days. Coding agents did most of the typing, under my architecture and review.
- Adversarial agent reviews before every merge, with verification agents reproducing findings empirically before fixes land
- Signed, notarized and self-updating. A release is one command that ships the DMG, changelog and fresh screenshots

### Flatsby

Collaborative household management app for flatmates with shared shopping lists and real-time sync across web and mobile.

- View Live: https://www.flatsby.com/
- View Code: https://github.com/RaphaelMitas/flatsby
- Tech Stack: Next.js, React, React Native, Expo, TypeScript, tRPC, Tailwind CSS, PostgreSQL, Drizzle ORM, shadcn/ui, Turborepo, OpenAI, Google Gemini, Anthropic, Effect-TS

Flatsby is a collaborative household management application for people sharing living spaces. It combines shopping lists, expense tracking, and AI-powered chat in a TypeScript monorepo with React Native mobile and Next.js (React) web clients. Features real-time synchronization, OAuth authentication, and group-based permissions.

- Cross-platform support with Next.js 15 web app and Expo/React Native mobile apps sharing code via Turborepo monorepo
- End-to-end type-safe APIs using tRPC v11 with real-time collaborative shopping lists
- OAuth authentication (Google, Apple) via Better Auth with PostgreSQL (Neon) and Drizzle ORM
- Modern UI with shadcn/ui, Radix UI, Tailwind CSS and NativeWind for consistent styling across platforms
- Expense tracking with splits for fair cost sharing among flatmates
- AI chat integration supporting OpenAI, Google Gemini, and Anthropic models for household assistance
- Effect-TS for robust error handling in the API layer

### Saytype

macOS menu bar app for push-to-talk speech-to-text — hold a key to record, release to transcribe and paste. Runs entirely offline on Apple Silicon.

- View Code: https://github.com/RaphaelMitas/saytype
- Tech Stack: Rust, Tauri, TypeScript, React, Python, Machine Learning

Saytype is a macOS menu bar application for push-to-talk speech-to-text. Hold the Right Command key to record, release to transcribe, and the text is automatically inserted at your cursor position. Built with a three-process architecture: a Tauri/Rust system integration layer handling hotkey capture (CGEventTap), native audio recording (CPAL at 16kHz), and clipboard-based text insertion; a Python sidecar running the Parakeet MLX model for offline transcription with the model kept hot in memory; and a React frontend for settings and onboarding. Requires no internet connection — all inference runs locally on Apple Silicon ML hardware.

- Fully offline speech-to-text using Parakeet MLX model on Apple Silicon — no cloud APIs required
- Three-process architecture: Tauri/Rust for system integration, Python sidecar for ML inference, React for UI
- Global push-to-talk via macOS CGEventTap with Right Command key detection on a dedicated CFRunLoop thread
- Native audio recording via CPAL with automatic sample rate conversion to 16kHz WAV
- Clipboard-based text insertion with save/restore and AppleScript keystroke simulation
- Sidecar keeps ML model loaded in memory between requests for fast sequential transcriptions via JSON IPC

### Shopping Assistant

AI-powered shopping assistant built for the Modern Stack Hackathon with real-time chat, web search, and usage tracking.

- View Live: https://shopping-assistant-wine.vercel.app
- View Code: https://github.com/RaphaelMitas/shopping-assistant
- Tech Stack: Next.js, TypeScript, Convex, OpenAI, Tailwind CSS, shadcn/ui, Vercel AI SDK

A full-stack AI shopping assistant built for the Modern Stack Hackathon (Sept 16 - Oct 1, 2025). Features agent-powered chat with GPT-5 and streaming responses, web search via Firecrawl, email authentication with Better-Auth, and usage tracking with Autumn. Built with Next.js 15, React 19, Convex, and Tailwind CSS 4. The hackathon featured a $25,000 prize pool.

- Agent-powered chat threads with GPT-5 and streaming responses using Convex Agent Threads
- Web search integration via Firecrawl with intelligent rate limiting
- Email-based authentication with Better-Auth including verification and password reset via Resend
- AI usage tracking and billing controls with Autumn for token management and feature gating
- Built for Modern Stack Hackathon (Sept-Oct 2025) integrating nine sponsor technologies
- AI agent with tool-calling capabilities using Vercel AI SDK for intelligent recommendations

## Skills & Technologies

- Languages: Swift, TypeScript, JavaScript, Python, Java, Rust, PL/SQL, Excel VBA
- Frameworks: Next.js, React, React Native, Expo, tRPC, Tailwind CSS, shadcn/ui, Drizzle ORM, Effect-TS, GraphQL, Express.js, FastAPI, Tauri, Material-UI, Plotly Dash, Oracle Apex, HTML, CSS
- Databases: PostgreSQL, Redis, Convex, SQL, Oracle
- Cloud: Kubernetes, Azure, Google Cloud
- Tools: Playwright, Cypress, Docker, Docker Swarm, GitLab CI, Turborepo, Grafana, Kafka, Knime
- AI: Agentic Coding, Claude Code, Cursor, Anthropic, OpenAI, Google Gemini, Vercel AI SDK, Machine Learning, Deep Learning
- Soft Skills: Problem Solving, Communication, Cross-functional Collaboration, Adaptability, Self-Learning, Mentoring

## Certifications

- Meta Front-End Developer Professional Certificate, Meta / Coursera, 2024-01 (https://coursera.org/verify/professional-cert/7RX9F5D9B443)
- Google Cloud Fundamentals: Core Infrastructure, Google Cloud / Coursera, 2023-12 (https://coursera.org/verify/46ZYZE9ULMFD)
- Generative Adversarial Networks (GANs) Specialization, DeepLearning.AI / Coursera, 2023-09 (https://coursera.org/verify/specialization/8JVE52C27DEA)

## Writing

- [I accidentally built the best Lightroom alternative](https://raphaelmitas.com/blog/i-accidentally-built-the-best-lightroom-alternative) (2026-08-17): How I built Photopipe, a free macOS photo culling and editing app, almost entirely by directing AI agents: a rebuild thrown away after a week, a working app in 48 hours, and where human taste still decides.
- [I let an agent move our Cypress suite to Playwright in a day, and you should too!](https://raphaelmitas.com/blog/cypress-to-playwright-with-agents) (2026-07-30): For most of the time I have been writing end-to-end tests, I would have told you Cypress was the nicer tool. The runner shows you the app as the test drives it. You can hover a command in the log and…
- Every post in full, with this CV: https://raphaelmitas.com/llms-full.txt