AI Product Engineering
Build AI features, web applications and internal tools. Take a prototype using real user data through to a production frontend and backend.
MVPs · Fullstack development · AI assistants · RAG and search
Germany / EU / US · Remote
Working with me
I take projects from problem to production.
I started with technology at 14, learning web security and SQL injection and building my first websites. That led to formal IT training, computer science studies and an award in an interdisciplinary university competition. Since 2015, I have worked professionally across infrastructure, data, web and product engineering; independently since 2018. Today I build AI products, integrations, data pipelines and cloud applications, and take on technical product ownership. I work directly with founders, business teams and engineers in German or English across Germany, the EU and the US.
- Experience
- In tech since age 14 · Independent since 2018
- Engagement
- Independent · Project or interim
- Focus
- AI · Integration · Data · Cloud · Product
- Region
- Remote · Germany, EU and US
Reference
“In our network, he has an outstanding reputation. Personally, I can say that he is a one-man army. He’s pragmatic, solves problems efficiently, and has the right experience.”
Thomas Marbella
Founder & CEO, OneCode GmbH
Original conversation
Tailorwine & Founders League
A conversation about Tailorwine and the work behind the digital product.
Watch on YouTubeIndependent AI Consultant & Product Engineer
AI consultingthat ships.
I build AI products, connect systems and data, and take ownership of technical product delivery.
AI product engineering, integration and automation, cloud operations, data engineering and technical product ownership. Available for defined projects or interim work within your team.
IndependentRemote Germany / EU / USGerman / English
Project briefdelivery
- Engagement
- AI consulting + delivery
- Scope
- Product · Integration · Data · Operations
- Outcome
- System in production
Definition of done
It runs in production. The owners understand it. The team can operate it.
ElectaCar
3 days
to a working first version
ElectaCar
6 weeks
from kickoff to production
Syntax Institut
3 months
from concept to first paying customers
VOXR
7 / 10
of the largest DAX companies use the product
The elephant in the room
AI demos are easy. Making one useful, reliable and worth the cost is the real work.
The model is rarely the bottleneck. Data, rules, permissions, integrations, failure modes and ownership are. Most of that stays invisible until something breaks.
What people see
Model, prompt, demo
What has to work
- Data and business rules
- Permissions and integrations
- Approvals and error paths
- Monitoring and ownership
That is the gap I close: from a convincing demo to a system people can rely on.
What you can hire me for
AI, software, data and product ownership.
Bring me in for a defined project or as part of your team. I can own a specific workstream or take a project from its first requirements through to production.
AI Integration & Automation
Connect AI agents to your data, APIs and existing systems. Replace manual work with traceable workflows, approvals and error handling.
Agent workflows · MCP · APIs and webhooks · System integration
AI Platform & Cloud Engineering
Deploy AI applications and keep them running reliably. Build background processing, deployment pipelines and monitoring so failures are visible and interrupted work can resume.
Deployment · Containers · Queues and retries · Monitoring
Data Engineering & Architecture
Connect scattered data sources and build reliable data models and pipelines. Give applications, analytics and AI usable data to work with.
SQL and PostgreSQL · Data pipelines · Data quality · Retrieval
Technical Product Ownership
Define product goals, work through requirements with users and stakeholders, and lead delivery. As a technical product owner or interim product partner, I own the roadmap, priorities and implementation.
Discovery · Roadmap and backlog · Acceptance criteria · Outcome measurement
When I am useful
Problems I can take on
- 01An AI use case or digital product needs someone to own both the scope and the build.
- 02Manual processes, scattered data or missing integrations are costing your team time.
- 03An existing system needs reliable operations or temporary technical product ownership.
How I work
An agent is not a business process.
Reliable automation needs clear rules, limited permissions and controlled handoffs. AI belongs only where fixed logic cannot do the job.
The result is not another demo chatbot. It is a traceable system that handles real work.
How a process becomes a systemRules, control, operations
- 01
Input
Data, request or manual process
- 02
Rules
Validation and fixed business logic
- 03
AI
Only where context and judgment are needed
- 04
Approval
Human control for critical steps
- 05
Operations
Logging, monitoring and clear ownership
API & dataLogs & error pathsHandover & operations
Selected projects
Production systems, not just features.
The difficult parts, the decisions behind them and the stack used to keep each system working in production.
2026
ElectaCar
Vehicle data, insurance and AI operations
Open project- Built the backend that combines government pricing, technical specifications, insurer mappings and media into one catalog for more than 100 vehicle models.
- Built a 12 stage publishing workflow that validates each vehicle, generates multilingual content and prepares production catalog images.
- Used Temporal for long running research, content and browser jobs, with retries and recovery after worker restarts.
- Separated the API, AI workers and browser workers from the public website, so unfinished or failed jobs never reach live pages.
- Built the quote and broker workflow with separate provider integrations, secure status access, email relaying and protected background processing.
Core stack
Python, FastAPI, Temporal, PostgreSQL, React, Next.js, TypeScript, Playwright and Docker.
2026
Heyamigo + Amigospace
First party AI agent runtime and context engine
Open project- Published a WhatsApp and Telegram AI assistant with more than 15,000 npm downloads.
- Runs Claude Code, OpenAI Codex, Gemini CLI and Grok Build through their first party command line tools, preserving each provider's sessions, reasoning and tool use.
- Keeps messages ordered in durable conversation queues and resumes interrupted work after a crash or restart.
- Runs browser tasks in isolated workers and separates memory into hot work, warm summaries and searchable cold source material.
Core stack
TypeScript, React, Node.js, Fastify, PostgreSQL with pgvector, SQLite, Playwright, MCP and Kubernetes.
2026
Google Ads Operations Platform
Controlled advertising and commerce operations
Private system
- Built a multi workspace platform that synchronizes live Google Ads structure, performance and account changes, then shows the difference before approved data is saved.
- Separated dry runs, Google validation and live publishing. Creating or adopting an ad group requires explicit confirmation before a campaign is changed.
- Kept AI generation outside the live publishing path. Every ad passes fixed content rules and operator review before it reaches Google.
- Connected Google Ads with Shopify products, orders and store revenue to calculate profitability by campaign, ad group and keyword.
- Recorded every publish attempt and Google account change, with historical snapshots that make unexpected changes traceable.
Core stack
TypeScript, Node.js, Express, Next.js, React, Google Ads API, Shopify Admin GraphQL, Zod, Zustand, Docker, Railway and Vitest.
2025 to 2026
RuntimeAffiliate
Multi merchant commerce and commission platform
Open project- Built one order and commission backend for multiple Shopify stores while keeping every store's data and balances separate.
- Tracks purchases through discount codes, Shopify App Pixel and a storefront SDK for standard Shopify themes and Hydrogen.
- Verifies browser purchase signals against the final Shopify order before recording commission, then adjusts balances when an order is refunded.
- Processes orders and refunds through a durable queue with duplicate protection, up to 10 retries and an hourly check for webhooks Shopify did not deliver.
Core stack
TypeScript, React, Next.js, PostgreSQL, Drizzle ORM, Shopify Admin API, AWS S3, Railway, Resend and Vitest.
2024 to 2026
BREAKEATS
Real time music and motion research
Private system
- Worked with a former NVIDIA manager and multiple patent holder on a new way to measure how closely movement follows music.
- Captured movement from iPhone and headphone sensors, then matched motion peaks against nine audio frequency bands in real time.
- Built latency aware scoring for hits, misses, silence, coverage and streaks across mobile and server systems.
- Moved recorded sessions into a Python worker for repeatable analysis and built diagnostic timelines for tracing errors in the matching algorithm.
Core stack
Swift, SwiftUI, CoreMotion, Accelerate/vDSP, LiveKit, Python, FastAPI, NumPy, SciPy, TypeScript, PostgreSQL, Docker and Kubernetes.
2024
DAYONE
FAMFAM music community platform
Open project- Developed a music community and event app cofounded by the Keinemusik DJ collective in a team of more than ten. It reached more than 12,000 users and the App Store Top 10 in Music.
- Built artist publishing, event invitations, reservations, push notifications and Stripe donations across the React Native app and React dashboard.
Core stack
React Native, Expo, React, TypeScript, AWS Cognito, API Gateway and Stripe.
2023 to 2024
Syntax Institut
Learning commerce, progress and certification
Open project- Built and launched the B2C learning platform from concept to paying customers in three months. It supports more than 3,000 active students for a training provider with over 100 instructors.
- Connected Stripe subscriptions and lifetime purchases to course access through signed webhooks, so confirmed payments unlock the correct content.
- Built server controlled progress, access and quizzes across modules and cohorts, with more than 27,000 quiz results processed in production.
- Added approval controlled PDF certificates and resumable learning across lectures, lessons and course materials.
Core stack
Next.js, React, TypeScript, NextAuth, Strapi, Node.js, PostgreSQL, Stripe, Sentry and React PDF.
2024
ÖGB Verlag
Adobe Experience Manager publishing platform
Private system
- Delivered the frontend migration of two large public websites to Adobe Experience Manager, including Gesunde Arbeit with more than 2,000 indexed pages.
- Built and adapted more than 30 components for navigation, search, forms, articles, galleries, downloads and audio.
- Built separate themes on a versioned shared platform, allowing several sites to reuse core components while keeping their own brands and layouts.
Core stack
Adobe Experience Manager, React, TypeScript, Tailwind CSS, SCSS, BEM and Webpack.
2022
Luxury Essence
Connected travel publishing platform
Open project- Built the publishing platform behind a luxury travel magazine with more than 170 articles, connecting editorial content directly with edel.travel hotels and tours.
- Connected the WordPress editor to the Directus travel catalog, keeping hotels, tours, destinations, advisors and media current without manual copying.
- Created 14 purpose built content blocks and limited the editor to approved layouts, giving the team publishing freedom without losing brand consistency.
Core stack
PHP, WordPress, Gutenberg, ACF, Directus, Tailwind CSS, JavaScript and Swiper.
2022
Tailorwine
B2B digital sommelier
Open project- Built the B2B MVP from scratch using existing consumer taste data. The working product helped Tailorwine win REWE and EDEKA.
- Turned sommelier knowledge into a 17 question flow that builds a taste profile and returns three wine recommendations in about 60 seconds.
- Built the experience for customer phones and in store touchscreens, without requiring customer accounts or a dedicated app.
Core stack
TypeScript, React, Tailwind CSS, Strapi and Sentry.
2017 to 2020
VOXR
High performance live event platform
Private system
- Led product and engineering for a platform used by the German federal government and 7 of the 10 largest DAX companies, including live events with more than 10,000 participants.
- Built a high performance version in pure JavaScript, keeping the client small and network requests minimal on overloaded venue WiFi.
- Designed live voting, questions, moderation and stage output for traffic bursts where thousands of people submit within seconds.
- Operated the system during live events and TV broadcasts, resolving critical failures while the platform was in use on stage.
Core stack
Pure JavaScript with a minimal browser runtime designed for unreliable venue networks.
Technical implementation
I can advise, architect and build.
If the bottleneck is the data model, API, frontend, agent logic or production setup, I can take it on directly. The stack serves the product, not the other way around.
AI systems & agents
Agent logic, tool calls, memory, browser automation and evaluation.
OpenAI · Anthropic · Grok · Tool Use
RAG & retrieval
Custom retrieval pipelines and framework-based orchestration for documents, knowledge and search.
LangChain · LangGraph · Haystack · RAGFlow
Web & product
Production-ready interfaces, admin flows and internal tools.
TypeScript · React · Next.js · Node.js
MCP & integrations
Custom MCP servers and connections for tools, data sources and production workflows.
MCP · REST · GraphQL · Webhooks
GEO & LLMO
AI-readable content, agent discovery and technical visibility in answer systems.
llms.txt · JSON-LD · Markdown · Agent Discovery
AI platform & cloud
Containers, deployments, durable queues, retries, error handling and monitoring for production AI applications.
Docker · Kubernetes · Temporal · CI/CD
Data engineering
Data models, imports, synchronization and verifiable pipelines for products, analytics and AI.
Python · SQL · PostgreSQL · pgvector
Technical product ownership
Turn product goals and user requirements into prioritized work, and coordinate delivery with business and engineering teams.
Discovery · Roadmap · Backlog · Acceptance criteria
What this changes
Architecture decisions are made against the live system, then implemented by the same person. No gap between the advice and the build.
A real project?
What would you like to build or improve?
Two or three sentences about your product, process, data or infrastructure, the outcome you need and the timeline are enough. I will tell you directly whether I can help and what I would do first.