CV
AI/ML Research Engineer
Multimodal AI · Computer Vision · Generative AI · On-Device Intelligence
AI/ML Research Engineer with a Ph.D. and more than five years of experience in multimodal AI, computer vision, image and video processing, efficient deep learning, and edge deployment. I develop AI systems across data preparation, model adaptation, training, evaluation, prototyping, and real-time inference.
My work combines research depth with practical software engineering: from vision-language models and spatio-temporal reasoning to RAG applications and optimized deployment on iPhone and NVIDIA Jetson. My research has appeared at ICLR, CVPR Findings, IEEE TCSVT, and IEEE ICIP, and I am a co-inventor on a U.S. patent.
Location: New Taipei, Taiwan
Work authorization: Authorized to work in Taiwan under a long-term open work permit.
Selected Impact
- Deployed a 31M-parameter multimodal model at 30-60 FPS on iPhone and NVIDIA Jetson Orin Nano.
- Reduced trainable parameters by 85% using parameter-efficient vision-language adapters.
- Built reproducible data-processing, training, and evaluation pipelines for more than 2.5 million video frames.
- Achieved a 5% ROC-AUC improvement through temporal-shift adaptation for video understanding.
- Achieved 99% verification reliability in a patented real-time audio-visual biometric system.
- Produced four peer-reviewed first-author papers and co-invented a U.S. patent.
- Mentored more than five graduate students and supported courses with over 40 students per class.
Work Experience
Research Assistant — Taiwan Tech (NTUST)
Taipei, Taiwan · Sep. 2020 - Feb. 2026
- Led a five-year research program in computer vision, vision-language adaptation, spatio-temporal reasoning, video understanding, knowledge distillation, model evaluation, and edge AI deployment.
- Proposed Temporally Shifted Distillation, allowing lightweight video models to learn predictive temporal representations from a frozen image-based teacher.
- Designed recurrent and linear-attention architectures for efficient long-range temporal reasoning.
- Developed parameter-efficient spatio-temporal adapters with 85% fewer trainable parameters than full fine-tuning while preserving multimodal alignment.
- Built reproducible processing, training, evaluation, and ablation pipelines for more than 2.5 million video frames with natural-language descriptions.
- Evaluated model quality and deployment readiness using ROC-AUC, mAP, accuracy, latency, model size, and FPS.
- Built end-to-end camera-perception prototypes and optimized inference through ONNX, TensorRT, CUDA, and CoreML.
- Deployed models on NVIDIA Jetson Orin Nano and iPhone at 30-60 FPS.
- Developed autonomous-vehicle perception systems with ITRI and Delta Electronics.
- Mentored more than five graduate students in research planning, implementation, experimental design, evaluation, and code review.
Teaching Assistant — Taiwan Tech (NTUST)
Taipei, Taiwan · Feb. 2021 - Jan. 2026
- Supported courses in Large Language Models and Applications, Intelligent Video Surveillance Systems, and Databases.
- Guided classes of more than 40 students through practical AI and software-engineering assignments.
- Assisted students with model implementation, experiment design, debugging, and technical communication.
Research Engineer — NTIS Research Centre
Czech Republic · Jan. 2019 - Oct. 2020
- Co-invented a patented real-time audio-visual biometric system that achieved 99% verification reliability.
- Developed modular C++/OpenCV/dlib pipelines for face detection, feature extraction, biometric representation, and multimodal matching.
- Designed software components and APIs connecting the C++ computer-vision backend to a Node.js authentication service.
- Generated QR-encoded identity hashes for real-time person identification, authentication, and verification.
- Contributed to the resulting U.S. patent.
R&D Software Engineer — Palaxo International Ltd.
Czech Republic · Sep. 2017 - Sep. 2020
- Architected a production-oriented C++ and OpenCV image-processing framework for an enterprise document-workflow platform.
- Developed automated OCR-based text extraction for heterogeneous image-based documents.
- Implemented dynamic masking and anonymization of sensitive information.
- Integrated computer-vision modules into production workflows managing more than 10,000 documents across Europe and the GCC.
Software Engineer — Kerio Technologies, Inc.
Czech Republic · Jun. 2016 - May 2017
- Implemented geolocation-based security features in C++ for a proprietary embedded router platform.
- Contributed to device-level access protection, debugging, and embedded networking software.
Selected Projects
Real-Time Traffic Hazard Reasoning on Edge Devices
Multimodal AI · Computer Vision · Edge Deployment · 2025-2026
- Developed a 31M-parameter multi-task model for traffic-hazard anticipation, localization, and live risk scoring from dashcam video.
- Deployed the model through CoreML on iPhone and ONNX/TensorRT/CUDA on NVIDIA Jetson Orin Nano.
- Achieved 30-60 FPS frame-by-frame inference on edge hardware.
- Built an iOS safety dashboard with live risk visualization, hazard localization, and sound and vibration alerts.
- Validated the complete system through on-road testing in Taipei.
Generative AI Research Assistant with RAG
Generative AI · Semantic Search · Local Inference · 2025
- Built a LangChain-based workflow for scientific-paper discovery, downloading, indexing, retrieval, and grounded question answering.
- Indexed full-text papers in ChromaDB using Hugging Face embeddings for semantic passage retrieval.
- Served Gemma and Qwen models locally through Ollama.
- Developed a FastAPI backend and Streamlit interface for locally hosted document search and interactive research assistance.
Parameter-Efficient Vision-Language Model Adaptation
Vision-Language Models · Video Understanding · 2023-2025
- Designed adapter-based spatial, temporal, and textual modules for adapting pretrained image-language models to video.
- Reduced trainable parameters by 85% compared with full fine-tuning.
- Preserved vision-language alignment while adding spatio-temporal reasoning capabilities.
- Improved accident-anticipation accuracy and earliness across real-world dashcam datasets.
Cross-Domain Semantic Segmentation
Autonomous Perception · Domain Adaptation · 2021-2022
- Developed a generator-discriminator domain-adaptation method for autonomous-vehicle perception.
- Achieved the best mIoU across four Cityscapes-to-Cross-City targets, with relative improvements of up to 20%.
- Packaged the training and evaluation environment with Docker for reproducible deployment.
- Completed the project in collaboration with ITRI.
Facial Biometric Descriptors for Multimodal Authentication
Computer Vision · Biometrics · C++ · 2019-2020
- Co-developed a patented CNN-based biometric verification system.
- Combined facial and voice representations into multimodal identity fingerprints.
- Implemented C++ computer-vision components and a Node.js verification backend.
- Encoded identity hashes as QR codes for document signing and real-time authorization.
Technical Skills
- Programming and software engineering: Python, PyTorch, C/C++, OpenCV, Linux, Git, CI/CD, Docker, Slurm, FastAPI, Streamlit, Node.js, REST APIs.
- Computer vision: Image processing, image recognition, camera-based perception, video understanding, semantic segmentation, domain adaptation, anomaly detection, accident anticipation, hazard localization.
- Models and architectures: Vision Transformers, CNNs, CLIP, MobileCLIP, V-JEPA, vision-language models, recurrent attention, linear attention, spatio-temporal modeling, hybrid architectures.
- Model development: Parameter-efficient fine-tuning, knowledge distillation, self-supervised learning, contrastive learning, foundation-model adaptation, training pipelines, evaluation pipelines, ablation studies, and model selection.
- Generative and agentic AI: RAG, semantic search, LangChain, LangGraph, Hugging Face, ChromaDB, FAISS, Ollama, MCP, Gemma, and Qwen.
- Deployment and optimization: ONNX, TensorRT, CUDA, CoreML, FP16, BF16, NVIDIA Jetson, iOS, GPU inference, distributed training, and real-time optimization.
Education
Ph.D. in Electrical and Computer Engineering — Taiwan Tech (NTUST)
Taipei, Taiwan · Sep. 2020 - Feb. 2026 · GPA: 4.3/4.3
Research area: Computer Vision and Deep Learning
Dissertation: Efficient Spatio-Temporal Deep Learning for Real-Time Video-Based Accident Anticipation on Edge Devices
Advisors: Prof. Yie-Tarng Chen and Prof. Wen-Hsien Fang
M.Sc. in Software Engineering — University of West Bohemia
Czech Republic · Sep. 2017 - Jun. 2020
Thesis: Information Extraction from Heterogeneous Image-Based Documents Using Templates
Result: Graduated with honours.
Master’s Exchange Programme — Taiwan Tech (NTUST)
Taipei, Taiwan · Sep. 2018 - Feb. 2019
Selected coursework: Intelligent Video Surveillance Systems, Intelligent Control Systems, and Basic Oral Chinese.
B.Sc. in Software Engineering — University of West Bohemia
Czech Republic · Sep. 2014 - Jun. 2017
Publications and Patent
Lightweight Spatio-Temporal Modeling via Temporally Shifted Distillation for Real-Time Accident Anticipation
P. Patera, Y.-T. Chen, W.-H. Fang, "Lightweight Spatio-Temporal Modeling via Temporally Shifted Distillation for Real-Time Accident Anticipation,"The Fourteenth International Conference on Learning Representations (ICLR), Rio de Janeiro, Brazil, 2026, pp. 1-20, url: https://openreview.net/forum?id=8zzfTSVds2
System and method for identification, authentication, and verification of a person based upon a short audio-visual recording of the person
K. Ekštein, M. Konopík, F. Pártl, P. Patera, "System and method for identification, authentication, and verification of a person based upon a short audio-visual recording of the person", US patent 2025/0184148.
A Multi-modal Architecture with Spatio-Temporal-Text Adaptation for Video-based Traffic Accident Anticipation
P. Patera, Y. -T. Chen and W. -H. Fang, "A Multi-Modal Architecture With Spatio-Temporal-Text Adaptation for Video-Based Traffic Accident Anticipation," in IEEE Transactions on Circuits and Systems for Video Technology, vol. 35, no. 9, pp. 8989-9002, Sept. 2025, doi: 10.1109/TCSVT.2025.3552895.
Spatio-Temporal Adaptation with Dilated Neighbourhood Attention for Accident Anticipation
P. Patera, Y. -T. Chen and W. -H. Fang, "Spatio-Temporal Adaptation With Dilated Neighbourhood Attention For Accident Anticipation," 2024 IEEE International Conference on Image Processing (ICIP), Abu Dhabi, United Arab Emirates, 2024, pp. 2452-2458, doi: 10.1109/ICIP51287.2024.10647316.
Teaching
Service, Awards, and Languages
Professional Service
- Reviewer, NeurIPS 2026.
Awards
- IT SPY 2020 — top-tier ranking among more than 1,400 master’s theses submitted across 14 Central European universities.
- M.Sc. Diploma with Honours.
- Taiwan Ministry of Education Scholarship, 2020-2024.
- English as a Medium of Instruction Teaching Certificate.
Languages
- Czech — native.
- English — fluent.
- Slovak — fluent.
- Mandarin Chinese — basic.