• Providing fractional AI and engineering leadership to early-stage and growth-stage companies, designing and shipping production-grade AI systems across media, process automation, and language-services sectors. 
• Engagements span the full delivery lifecycle ? from technical strategy, architecture, and team mentoring through hands-on implementation, deployment, and MLOps -turning state-of-the-art AI research into reliable, cost-effective systems that run in production.
• Typical work combines large language models, retrieval-augmented generation, and classical data engineering, with a strong emphasis on evaluation, observability, and total cost of ownership.

Customer: Television. AI GmbH, Frankfurt

Tasks:

• Advising on and building AI-powered video editing and production pipelines for newsroom content. Designed the end-to-end system covering media ingestion, transcription, entity identification, and video semantic-knowledge extraction, and built fine-tuning pipelines that capture the distinct editorial voice of individual TV channels. 
• Added human-in-the-loop review and evaluation to keep quality consistent at scale, powering AI video-editing tools now adopted by several of Germany?s leading public broadcasters.

Customer: Traducciones 24/7, Bogota, Colombia 

Tasks:

• Building automation for online quoting and document-processing workflows to support certified translation delivery. 
• Automated document intake, layout-aware text extraction, and price estimation, cutting manual quoting effort and shortening turnaround time for the translation team.

Customer: Seneca Evolution UG, Lorsch

Tasks:

• Directed the technical strategy behind the company?s internal financial-controlling product, combining ETL pipelines with retrieval-augmented generation (RAG) to build a per-customer knowledge base powering AI-driven search and automated financial controlling tools.
• Defined the data model and ingestion strategy so new customers could be onboarded with minimal manual configuration.

Customer: Audioscrape Inc., San Francisco, US

Tasks:

• Contributed to the internal audio-ingestion pipeline, scaling data-collection infrastructure to reliably source audio content across a wide range of internet platforms, many with strong anti-bot protections, feeding a searchable index of over a million hours of podcasts and recordings. Focused on resilience, throughput, and cost control of the crawling and processing pipeline.
• Advising founders and leadership on AI roadmap, build-vs-buy decisions, model and tooling selection, architecture reviews, and hiring for technical teams.

valuation 18 months after founding. My departments were responsible for maintaining and developing all the technical infrastructure that sustains more than 100 million euros in revenue every year for more than 100,000 products on Amazon, eBay, Walmart, and more across four continents. This includes critical infrastructure affecting the P&L directly. I set the technical vision and roadmap, owned reliability and cost of the production estate, and led a cross-functional organization spanning machine learning, data engineering, and platform/software teams.

• Set technical direction, standards, and delivery processes for a cross-functional organization covering ML, data engineering, and platform teams, including hiring, mentoring, and performance management.
• Architected and operated mission-critical systems enabling pricing, listings, ads, inventory, and finance across global marketplaces, treating them as always-on services with clear ownership and SLAs.
• Architected a centralized internal tech hub ? a single platform for product data and internal tooling ? used by roughly 50% of the company.
• Maintained Airflow jobs for 5k+ pages scraped daily, 2k+ invoices processed per execution, 50k+ data rows written daily, and 30+ marketplaces processed for 100+ brands, backed by ETL pipelines and warehousing on Snowflake/dbt.
• Automated 70?80% of demand planning (replacing a fully manual, four-planner process), cutting forecast error (MAPE) by roughly 30% (0.4 to 0.3), with the largest gains on long-tail, slow-moving SKUs where error was roughly halved.
• Reduced overstock costs by an estimated 700K+ euros annually (a conservative 3?4% improvement on a 15?20M euro overstock-fee base) through better forecasting, while freeing planner time for higher-value work.
• Strengthened reliability and delivery speed by standardizing CI/CD, monitoring, and infrastructure-as-code across services running on Kubernetes (AWS and Azure).
• Drove company-wide adoption of generative AI, embedding LLM-powered automation and internal copilots into everyday operational workflows.

Led the ML team and all ML-related development at the company. Worked closely with the largest automotive OEMs in Germany to deliver key POCs and milestones, balancing research ambition with the reliability and timelines expected by automotive partners.

• Defined the company?s ML strategy and roadmap, coordinating research and engineering to meet OEM milestones on time.
• Delivered the unknown-object detection (safety system) showcased in Munich for the IAA demo together with Mercedes-Benz, Volkswagen, BMW, Jaguar, and Ford ? a real-time early warning function with explicit detection, threshold, and triggering logic.
• Defined the operational design domain (ODD) and system architecture for camera based perception functions ? inputs, outputs, latency and accuracy targets, and failure/degradation behaviour.
• Led important POC projects for several of the biggest car manufacturers in Germany and the world, including production lines for popular car models.
• Owned the ML processes end to end ? 2D/3D detection, depth, segmentation, and model design ? and set the technical direction for 3D understanding.
• Led dataset construction, annotation guidelines, and data-collection strategy for diverse and adverse conditions (low light, glare, weather, occlusion, degraded and cluttered scenes) across OEM programmes in Germany, Korea, and the US.
• Established the training, evaluation, and deployment workflow that the ML team standardized on, improving consistency and reproducibility.
• Mentored engineers and grew the team?s capability across perception and deep learning.