Mateusz Kędzia
Beijing Nievedor Intelligent Technology Co., Ltd.
Contributing to AI innovation in a rapidly growing company, responsible for AI product development, technical content creation, and strategic partnerships. Built RAG systems, summarization engines, and multi-provider LLM integrations (OpenAI/GPT-4, Anthropic/Claude, Google/Gemini) directly and via OpenRouter. Designed and deployed n8n-based AI automation pipelines for business process optimization. Created fully functional demos and video tutorials showcasing latest AI technologies (Sora2, NanoBanana, LibreChat, Cursor) published on Bilibili, Xiaohongshu, and WeChat. Led a collaboration with DiDi, integrating DiDi MCP and Gaode MCP into a custom LibreChat deployment, enabling seamless access to mobility and mapping services via conversational AI. Managed all cloud infrastructure using Terraform (IaC) on AWS. Delivered multi-provider LLM integrations enabling cost-effective model routing and published AI tech showcases reaching audiences across major Chinese social platforms. Successfully shipped DiDi×Gaode MCP integration as a production-ready conversational interface and established scalable, reproducible cloud infrastructure via Terraform/AWS.
Vrije Universiteit Amsterdam (with TNO & Arnhems Buiten)
Part of the EU-funded HEDGE-IoT project deploying smart meters, V2G charging stations, residential batteries, heat pumps, and solar PVs across 15 buildings in Arnhem, Netherlands, coordinated through Building and Energy Management Systems (BEMS/EMS). Designed the IoT data science platform, implemented semantic data pipelines using SAREF ontologies, and built ML workflows for real-time energy optimization. Built and deployed a Dockerized JupyterHub platform supporting 50+ users, developed semantic adapters and data conversion pipelines into knowledge graphs using RDF/SAREF, and implemented edge-cloud IoT architecture enabling real-time anomaly detection and predictive energy scheduling. Integrated Knowledge Engine (KE) with explainable AI for cross-building energy flexibility operations. The platform enabled real-time monitoring, semantic integration of distributed energy assets, and grid fault detection across 15 buildings, and was successfully showcased at the HEDGE-IoT conference, supporting 50+ researchers in flexibility services and predictive optimization.
University of Amsterdam
Developed an efficient web crawler for automated data extraction and structuring to support large-scale research requiring high-integrity news datasets. Implemented anti-blocking and relevance filtering algorithms, automating the processing of over 1 million news articles and achieving 99% data integrity, providing reliable datasets for research.
Vrije Universiteit Amsterdam & Beijing University of Technology
Background: Trajectory prediction models benefit from cross-task knowledge transfer but often lack efficient distillation methods.
Task: Develop a knowledge distillation framework achieving SOTA performance in map-matched trajectory prediction through representation alignment.
Action: Designed and implemented cross-task distillation models with representation alignment techniques, conducting comprehensive benchmark evaluations.
Result: Matched SOTA performance through efficient representation alignment. Achieved thesis grade of 8/10 with overall Master's average of 8/10.
Reflection: Deepened expertise in knowledge distillation, trajectory prediction, and cross-task transfer learning.
Vrije Universiteit Amsterdam
Background: Integrating multi-source IoT data requires transparent, interoperable, and explainable AI solutions.
Task: Design explainable models and data pipelines to unify heterogeneous IoT data.
Action: Developed device behavior prediction models, implemented data integration based on RDF and SPARQL, ensuring model explainability.
Result: Delivered robust IoT data integration and prediction system, achieved 8.0 grade, demonstrating practical value of explainable AI and semantic data integration.
Reflection: Strengthened foundations in semantic data integration, time series prediction, and AI transparency.
Led practical courses and guided 10+ students in machine learning and data projects.