CV

• Engineered data ingestion and processing pipelines for DHL’s Document AI, integrating Gemini through the enterprise GenAI Gateway and using rasterized pages to improve document extraction.
• Built Grafana dashboards to monitor API costs, token usage, and errors across AI services, supporting cost optimization and operational troubleshooting.
• Developed comprehensive Power BI dashboards to monitor conversational analytics, automated resolution rates, and user-flow performance for the Deutsche Post & Parcel AI voice bot (Jana).
• Built a Pydantic AI multi-agent system for DHL’s internal-control validation, with Python orchestration, deterministic compliance checks, and agent tool calling for approved rule and template retrieval.
• Developed a GenAI React-based web application for summarizing feedback, including an interactive frontend and structured AI-generated recommendations for DHL Consulting.
• Designed the backend with MongoDB and RabbitMQ for the DHL Express Customs AI Agent.

• Built and deployed a domain-specific RAG chatbot for aircraft technical manuals, optimizing retrieval precision through hybrid search and semantic chunking.
• Built a centralized Power BI dashboard tracking operational KPIs for more than 2,000 consultants. Automated data ingestion with Power Query M, reducing reporting time by 80%.
• Deployed scalable LLM applications using Azure AI Foundry and Copilot Studio, integrating agentic workflows to automate business processes.
• Engineered a containerized GenAI project-management platform on Azure, using context-aware prompts to populate complex templates from uploaded technical documentation and reducing project initialization time by up to 60%.
• Developed Microsoft Power Platform (Power Automate and Power Apps) solutions for cross-functional digital-transformation initiatives, collaborating with HR, Finance, and Project Directors to align business requirements and project deliverables.
• Built ETL and ELT pipelines using Microsoft Fabric Dataflows and Databricks to support analytical data models for 10+ dashboards.
• Coordinated user acceptance testing with Project Directors across more than three German regions for a cost-center automation project.

• Developed and evaluated regression models in Python using scikit-learn for sales forecasting.
• Built a supplier recommendation system using collaborative filtering to reduce churn rate.

• Master’s Thesis: “From Reactive Control to Holistic Planning: Evolutionary Prompt Optimization for Hybrid Deterministic-Neural Agents in Long-Horizon Environments” — open-source Python agentic LLM framework with a deterministic engine, constrained decoding, and evolutionary prompt optimization.
• Focus areas: Machine Learning, Neural Networks, LLMs, AI Agents and Foundation Models

• Bachelor’s Thesis (Grade: 1.3): “Deepstory: A New Storytelling Experience”
• Double Degree, Focus areas: Machine Learning, UX/UI, HCI, Visual Communication and AI Ethics

A free bilingual AI platform that helps users access and understand Hong Kong legal information.

• Built and shipped a full-stack legal AI chatbot used by the public, answering more than 20,000 questions and owning the RAG pipeline design, backend, frontend, and deployment.
• Implemented hybrid-search RAG combining dense retrieval and BM25 sparse retrieval across more than 20 years of Hong Kong Judiciary data to optimize retrieval accuracy for bilingual legal queries.
• Implemented an LLM-as-a-judge evaluation and guardrail workflow, achieving an 80% pass rate across 100 test queries against defined factuality and response-quality criteria.

AI storytelling system using GPT-2 and a custom-trained text-to-speech model to generate videos.

• Engineered an end-to-end real-time inference pipeline, serving a fine-tuned GPT-2 model and a custom PyTorch TTS model through a production Flask web app.
• Built a custom audio-preprocessing pipeline for segmentation and silence normalization.
• Trained DC-TTS convolution models in PyTorch using preprocessed audio and transcriptions.