Gergely Papp CV (PDF)
Gergely Papp standing on a rock ledge at the Grand Canyon

Gergely Papp

Sr. Machine Learning Engineer

Computer Vision and Language Processing

Amsterdam, NL

Download CV

Experience

  1. Data Scientist & ML Engineer

    Datacation B.V. 01/2026 – present

    • Design and train vision systems for client projects: data flow, label definitions, detection, classification and 3D segmentation models, post-processing
    • Set up on-site cameras, edge devices and data collection for vision deployments
    • Engineering lead on an LLM / RAG project: own the backlog and coordinate the team
    • Fine-tune LLMs (LoRA) and embedding models
    • Deploy to Azure, the in-house GPU server and edge / on-prem client hardware; optimise inference (ONNX, TensorRT, quantisation)
    • Built CI/CD for continuous model training and validation
    • Scope PoCs and budgets with clients; weekly updates and result presentations
    • Admin of the team GPU server
    • Interview and onboard data scientists

    PyTorch · Hugging Face · LangChain · Azure · Docker · GitHub Actions · ONNX · TensorRT

  2. AI Engineer

    Freelance (TruthWorks) 06/2025 – 01/2026

    • Built the OCR pipeline for case documents and an image-captioning system for visual evidence
    • Designed RAG workflows and knowledge graphs over investigation bundles of thousands of pages
    • Built summarisation and contradiction detection across case files for US police departments

    AWS Bedrock · LangChain · Docker · PostgreSQL · S3 · Redis

  3. Machine Learning Consultant part-time, 10/2020 – 07/2024

    Asura Technologies

    • Lead ML engineer in a team of ~5 on computer vision projects
    • Raised object detection accuracy from 75% to 93% by debugging the model re-implementation, reworking the training recipe and post-processing
    • Advised on technology choices and implementation to streamline the team's workflows

    Machine Learning Engineer 10/2018 – 10/2020

    • Designed, trained and served real-time object detection models
    • Built the in-house license plate recognition (ALPR) and OCR engine from scratch, replacing a paid third-party solution
    • Built an automated parking system tracking vehicles across a car park
    • Delivered PoC models for new customers

    TensorFlow · OpenCV · Scikit-Learn · Django · Flask · C# · Docker

  4. Deep Learning Research Engineer

    Alfréd Rényi Institute of Mathematics 10/2020 – 07/2024 · part-time from 09/2022

    • Researched representation learning (self-supervised image classification, similarity of neural network representations, NLP transformers) and generative models (GANs, VAEs)
    • Designed custom architectures, layers and loss functions in PyTorch; multi-GPU distributed training (DDP)
    • Built scalable pipelines for model training and evaluation
    • Co-authored the first NeurIPS paper from a Hungarian institute

    PyTorch · Hugging Face · Wandb · Docker · SLURM

  5. Risk Analyst (AI team)

    Morgan Stanley 08/2017 – 05/2018

    • Automated data-processing workflows, saving ~100 hours per week
    • Built clustering and forecasting models on tabular data
    • Built a hierarchical-clustering tool to reconcile balance sheets across accounting systems, and an OCR system extracting structured data from invoices

    Python · Pandas · Scikit-Learn · Spark · SQL · Q

Selected projects

  • Datacation · Computer vision engineer · 2026

    KLM Cargo

    A visual AI system supporting KLM Cargo's cargo registration pipeline. Ordered and installed the warehouse cameras, connected them to a local edge device, set up data collection, and trained and evaluated detection, classification and 3D segmentation models; kept in contact with the client and reported progress.

    PyTorch · Edge deployment · Detection · Classification · 3D segmentation

  • Datacation · Engineering lead · Ongoing

    LLMs4EU

    Building a RAG-based search and a democratic tourism recommendation chatbot for the EU. Leading the engineering work: owning the backlog, coordinating the team members' work, and driving improvements in embedding-based retrieval.

    PyTorch · Hugging Face · LangChain / LlamaIndex · Vector database · Azure OpenAI / cloud LLM APIs

  • Asura Technologies · ML Engineer

    License Plate Recognition

    Initiated and built a ViT-based license plate recognition system that successfully replaced an expensive third-party solution.

    Python · TensorFlow · OpenCV · NumPy · C#

  • Asura Technologies · ML Engineer

    Home Quarantine System

    Created an ML system to track quarantine compliance during COVID-19 pandemic, with face detection and anti-cheating mechanisms. Adopted by Hungarian government and used by tens of thousands of citizens.

    Python · TensorFlow · OpenCV · NumPy

    Product page

  • Alfréd Rényi Institute · Research Engineer · NeurIPS 2021

    Functional Similarity

    Demonstrated that geometric and functional similarity in neural networks are distinct concepts using affine transformations between networks. First NeurIPS paper accepted from a Hungarian institute.

    Python · PyTorch

    GitHub

  • University of Amsterdam · First author

    Stitch-BERT

    Analyzed how NLP transformers fine-tuned for different languages and tasks relate geometrically and functionally, revealing potential for cross-task insights.

    PyTorch · Python

    GitHub

  • Alfréd Rényi Institute · Research Engineer

    Self-Supervised Learning

    Explored innovative self-supervised image classification methods competing with state-of-the-art approaches, achieving promising results on smaller datasets.

    Python · PyTorch · Wandb · Docker

    GitHub

  • University of Amsterdam · MSc AI

    MSc Thesis: Cross-Taxonomy Generalization

    Investigated Vision Transformers' ability to generalize across object properties (shape, texture, color, count) on CLEVR-4. Introduced Cross-Taxonomy Generalization (CTG), a novel problem in classification as a subcategory of Few-Shot Learning.

    Python · Hugging Face

  • Freelance · AI Engineer

    TruthWorks

    AI-powered investigation platform for US police departments featuring OCR, image captioning, and RAG workflows over thousands of case-file pages.

    AWS Bedrock