← Back to Portfolio

PATRICK FELBER

AI Engineer

Bad Vöslau, Austria•patrick.felber@gmx.at•+43 664 5855876

Portfolio & Projects: patrickfelber.com

Professional Summary

AI Engineer with 10 years at Springtime Technologies, where I built and own the core automation and extraction capabilities powering Invoicetrack for Fortune 500 clients. Deep expertise in building and integrating production AI systems (7 years), from CNNs (2019) to modern LLMs/VLMs. Expert in integrating computer vision and NLP into scalable systems processing 1.2M+ documents monthly. First to introduce AI at company (2019), leading successful integrations that achieved 30-70% error reduction and 50% reduction in support workload. Proven track record architecting end-to-end ML systems on Azure GPU infrastructure serving enterprise clients including Boehringer Ingelheim, Bosch, DHL, and Evonik.

→ See detailed project portfolio with demos, metrics, and technical deep-dives at patrickfelber.com

Technical Skills

AI/ML:PyTorch, TensorFlow, HuggingFace Transformers, LLMs/VLMs, RAG, Langchain, vLLM, Computer Vision, NLP, Model Training & Evaluation, CNNs, Donut, LayoutLM, Azure OpenAI Service, Azure Document Intelligence, Azure Content Understanding
Languages:Python, C#, TypeScript, SQL
Infrastructure:Azure GPU VMs (A100), Docker, Kubernetes, Terraform, CI/CD (TeamCity, Octopus Deploy)
Databases & Search:Weaviate (Vector DB), Elastic Search, SQL Server, Semantic Search
Frameworks:FastAPI, ASP.NET, React, Next.js, .NET Framework → .NET 10

Professional Experience

AI Engineer & Technical Lead

2019 - Present

Springtime Technologies GmbH

Vienna, Austria
  • Led VLM-based document extraction system serving 1.2M+ multi-page documents monthly, achieving 30-70% error reduction across fields and higher automation rate than manual clerks by architecting self-validation training pipeline on Azure 2x A100 GPUs
  • Managed team of 3 engineers (2 backend, 1 frontend) developing production LLM/VLM pipeline using vLLM, Docker, Kubernetes, and Terraform, delivering system to all enterprise customers in 2025
  • Architected company-wide RAG chatbot using Langchain (early adoption, v0.x), Weaviate vector database, and multi-modal embeddings, reducing Development team support requests by 50% through semantic search vs traditional keyword search
  • Championed AI coding tools adoption, evaluating GitHub Copilot, Cline, Cody, and Aider, presenting ROI analysis to leadership, and deploying GitHub Copilot to 35-40 employees (~50% of Austrian office including 25 developers)
  • Migrated entire OCR infrastructure from legacy provider to Azure Document Intelligence, avoiding 100% price increase (hundreds of thousands EUR monthly), achieving zero downtime, and future-proofing with Microsoft's AI-powered Content Understanding
  • Pioneered company's first AI integration by designing ResNet-based CNN with custom embedding layer for invoice field extraction, achieving 10% error reduction vs manual processing and establishing end-to-end ML training/deployment platform

Software Engineer

2016 - 2019

Springtime Technologies GmbH

Vienna, Austria
  • Built production template-based extraction system with custom layout matching algorithm that has maintained 0% failure rate since 2019 deployment, supporting multi-line table headers and complex invoice structures
  • Designed semantic rule engine with spatial relations (e.g., "date_label <left_of> date") and integrated Aho-Corasick algorithm for searching millions of PO numbers in OCR text within milliseconds
  • Developed Elastic Search recommendation system indexing historical documents to auto-populate invoice fields, reducing manual entry time by >50% through intelligent dropdown suggestions
  • Built document automation solutions for invoice processing platform serving enterprise clients including Boehringer Ingelheim, Bosch, DHL, Evonik, Festo, and Takeda

Software Developer (Internship - Bachelor Thesis)

2014 (3 months)

A1 Telekom Austria

Vienna, Austria
  • Automated quarterly customer satisfaction reporting for Austria's largest telecom provider, replacing manual spreadsheet-based workflow and eliminating multi-person team effort

Education

Master of Science in Computer Science

2012 - 2016

Fachhochschule Wiener Neustadt, Austria

Specialization: Software Architecture and Design

Thesis: 3D vs 2D Terrain Rendering Performance Optimization on Mobile Devices

Bachelor of Science in Computer Science

2008 - 2012

Fachhochschule Wiener Neustadt, Austria

Focus: Software Engineering

Thesis: Automated Customer Satisfaction Reporting System (A1 Telekom Austria)

Key Projects & Achievements

  • AI Technology Early Adoption Timeline: Neural Networks (2014) → GANs (2015) → Reinforcement Learning (2017) → CNNs for Production (2019) → Langchain/RAG (2022) → LLMs/VLMs (2023-2026)
  • Built profitable reinforcement learning trading bot (2017-2019) using OpenAI's PPO algorithm, custom Gym environment with visual market representation, deployed on live BitMEX exchange with real capital
  • Developed computer vision poker bot (2014) with screenshot-based game state extraction, Monte Carlo hand strength calculation, custom HUD, and 100% card recognition accuracy using pixel-perfect detection
  • Engineered custom 3D game engine (2011-2012) in C# using XNA/DirectX 9, implementing TCP hole punching for peer-to-peer multiplayer without dedicated servers, shadows, physics, and custom shaders

Additional Information

Languages: German (Native), English (Fluent - daily professional use)

Interests: Table Tennis (33 years competitive), Technology Research