CV

CV of A.D (David) Le — AI Researcher & Engineer.

Contact Information

Name Le Anh Duy
Professional Title AI Researcher & Engineer
Email leanhduy497@gmail.com
Phone (+84) 349749495

Professional Summary

My research began with recognizing content and two-dimensional structure in documents, and my current document VLM work extends that foundation in multimodal understanding. In one-shot handwriting generation, I learned discrete writer-style concepts from a single reference image to guide text-conditioned diffusion, motivating my interest in composing visual concepts from few examples. I want to build generative models that use vision-language understanding to interpret a creator’s intent from language, sketches, reference images, and iterative feedback. My PhD goal is personalized image and video generation with control over composition, appearance, and motion while preserving identity and temporal consistency.

Experience

  • 2025 -

    Remote from Ho Chi Minh City, Vietnam

    AI Engineer
    Drimco AI
    Build document AI workflows for technical document processing, including core document structure and layout analysis models.
    • Developed a layout analysis solution for customer technical documents, combining an M2Doc-inspired multimodal detector with rule-based post-processing; achieved 0.90 mAP on an internal evaluation set of customer documents.
    • Inspected predictions and analyzed errors in difficult layouts; experimented with fine-tuning document VLMs, including MinerU2.5 and dots.ocr, for layout analysis.
    • Language/Framework: Python, PyTorch, FastAPI, MLflow, Docker
  • 2024 -

    Ho Chi Minh City, Vietnam

    Senior AI Engineer
    Viettel AI
    Led OCR projects for major insurance and banking clients. Core R&D member researching document understanding and generative modeling.
    • For a Vietnamese document VLM project, built training-data pipelines: deployed a tagging system across 500,000+ real images and produced approximately 500,000 synthetic documents with parsing and layout annotations.
    • Fine-tuned and evaluated EAGLE-3 and DFlash speculative decoders on internal document data; the best setup achieved 4.2× lower latency than baseline Qwen3-VL-4B on B200, with unchanged internal OmniDocBench-style overall score.
    • Led a three-person team developing text recognition models on 10M+ samples for banking, insurance, and legal documents; achieved 98% accuracy on an internal evaluation set covering scans and camera captures, and delivered OCR systems through production deployment.
    • Core R&D member of the team, researching new technologies in document understanding and generative modeling, applying findings to improve company products, and publishing research papers at top-tier conferences (WACV, ICDAR).
    • Mentor junior engineers on research methodology, experiment design, and engineering best practices.
    • Drive technical decisions on model architecture and system design, collaborating with product and business teams to translate requirements into scalable AI solutions.
    • Language/Framework: Python, PyTorch, FastAPI, Wandb, LaTeX, Docker, CI/CD
  • 2021 - 2024

    Ho Chi Minh City, Vietnam

    Junior AI Engineer
    Viettel AI
    Developed OCR/eKYC systems for government and business clients.
    • Participated in developing OCR/EKYC systems to automate information extraction from various document types.
    • Researched and applied SOTA architectures for text recognition, information extraction, and layout analysis.
    • Language/Framework: Python, PyTorch, Docker
  • 2020 - 2021

    Ho Chi Minh City, Vietnam

    Junior AI Engineer
    ARI Technology Joint Stock Company
    Built AI systems for customers including face recognition, object detection, and AI chatbot.
    • Language/Framework: Python, PyTorch, RaSa, Flask
  • 2019 - 2020

    Ho Chi Minh City, Vietnam

    Member of Academy Team
    BK Artificial Intelligence Club
    Organized seminars and workshops on Machine Learning and Deep Learning topics.
  • 2019 - 2019

    Ho Chi Minh City, Vietnam

    Teacher Assistant
    Faculty of Computer Science & Engineering, HCMUT
    Assisted the lecturer in the Data Structures and Algorithms course by preparing exam questions, grading assignments, and providing tutoring support to undergraduate students.
  • 2018 - 2018

    Ho Chi Minh City, Vietnam

    Volunteer
    Ho Chi Minh City 'Green Summer' Campaign
    Participated in the city’s annual youth volunteer campaign.

Education

  • 2021 - 2023

    Ho Chi Minh City, Vietnam

    Master of Engineering in Computer Science
    VNU-HCM University of Technology
    Computer Science
    • Cumulative GPA: 8.29/10.00
    • Thesis: A system for extracting mathematical expressions from document images
    • Thesis Grade: 9.94/10.00
  • 2017 - 2021

    Ho Chi Minh City, Vietnam

    Engineer of Computer Science
    VNU-HCM University of Technology
    Computer Science
    • Cumulative GPA: 8.67/10.00
    • Thesis: A study on a diagnosis system for Eye diseases
    • Thesis Grade: 9.62/10.00

Projects

  • CONSTANT: One-Shot Handwriting Generation (Apr. 2024 – Apr. 2025)

    Viettel AI; advisor: Tuan-Anh Tran; first-author research

    • Addressed the difficulty of recovering a new writer’s style from a single, potentially noisy sample. Proposed Style-Aware Quantization, which learns discrete visual style concepts instead of relying on fixed high-frequency features, with a contrastive objective to separate writers.
    • Introduced a multiscale latent patch contrastive objective that aligns generated and target patches during diffusion training, sharpening local stroke details in a single-stage framework.
    • On the IAM test set, SAQ reduced FID from 16.73 to 12.47; the full model reached FID 10.20 / HWD 0.74, respectively 36% / 30% lower than One-DM. Evaluated on English, Chinese, and Vietnamese handwriting.
  • Hybrid Vision Transformer for Mathematical Expression Recognition (Jun. 2022 – Jun. 2023)

    HCMUT; advisor: Tuan-Anh Tran; master’s research

    • Addressed long-range symbol relationships and missed or repeated symbols in formula parsing using a hybrid CNN–ViT encoder with 2D positional encoding and a coverage-attention decoder; achieved 89.94 BLEU-4 and 86.48% image exact match without spaces on IM2LATEX-100K.
    • Extended the paper in my master’s thesis by designing LIMD, a 1,564,755-pair image–LaTeX dataset, and building its collection, normalization, rendering, and outlier-filtering pipeline. [Thesis]

Publications

  • 2026
    CONSTANT: Towards High-Quality One-Shot Handwriting Generation with Patch Contrastive Enhancement and Style-Aware Quantization
    WACV 2026 (Oral, Award Finalist)
  • 2023
    Formerge: Recover Spanning Cells in Complex Table Structure Using Transformer Network
    ICDAR 2023 (Poster)
  • 2022
    A Hybrid Vision Transformer Approach for Mathematical Expression Recognition
    DICTA 2022 (Oral)

Awards

  • 2026
    Award Finalist, WACV 2026
    IEEE/CVF WACV
  • 2025
    Recognition of individuals for excellent achievements in Science and Technology activities
    Viettel AI
  • 2024
    Runner up at ICDAR Competition on Artistic Text Recognition
    ICDAR
  • 2023
    Runner up at Automatic Speech Recognition tasks - Viettel Hearted AI Challenges
    Viettel
  • 2022
    Recognition of individuals for excellent achievements in Science and Technology activities
    Viettel AI
  • 2021
    Top 9/40 AICovidVN 115M Challenge: Covid Cough Detection Challenge (Final Round)
    AICovidVN
  • 2019
    Student of Five Merits Scholarship
    VNU-HCM University of Technology
  • 2018–2019
    Academic Encouragement Scholarship
    VNU-HCM University of Technology

Certificates

Skills

Programming Languages: Python, C/C++, Java
Technologies: Diffusion models, vision-language models, PyTorch, vLLM, MLflow, TensorFlow, Google Cloud Platform, OpenCV, NumPy, Scikit-learn, Pandas, Matplotlib, PyQT5, Wandb, FastAPI, Git, Draw.io, Docker, LaTeX
Platforms: Linux, Windows

Languages

English : Professional working proficiency
Vietnamese : Native speaker