AI Engineer

Hoàng Phú Hào

AI Research Engineer

076******* **********@*****.*** District 7, HCMC

EDUCATION

Le Quy Don High School for the gifted – Ba RiaVung Tau Sep 2013 - May 2016

Bach Khoa University – HCMC Oct 2016 - Aug 2020

WORK EXPERIENCE

Cinnamon AI Nov 2019 - now

AI research engineer

Conduct researches:

-Work on Flax Scanner product - an intelligent document processing system to extract data from free-format document images.

-Develop the OCR, Layout Analysis & Information Extraction Engine for the Flax Scanner.

-Technical report and sharing, discuss, catch up other research.

Work on projects: Apply Flax Scanner's variants to projects.

-Current role: Project Technical Lead

-Project technical management (technical assessment, cross-functional team communication (delivery, AI, SE, QA), progress control, task assignment,…)

-Develop AI models.

-Solve related issues, ensure system performance.

-Customers: Daikin, Showadenko, EDSP...

ACTIVITIES & AWARDS

Zalo AI Challenge 2020 Nov 2020 - Dec 2020

participated as an individual

-Build traffic sign detection system, given a highly practical real-world dataset.

-Rank 19th out of more than 200 teams/individuals (https://challenge.zalo.ai/portal/traffic-sign- detection/leaderboard)

Mathematical Olympiad Jan 2016 - Feb 2016

participated as an individual

-First prize at Ba Ria Vung Tau Mathematical Olympiad, be a member of BRVT Team attending Vietnam Mathematical Olympiad 2016.

SKILLS

-Good knowledge of math.

-Strong Machine Learning fundamentals, good understanding of Deep Learning concepts, algorithms.

-Good knowledge of Computer Vision / Image Processing.

-Experienced with Graph Neural Network.

-Ability to quickly read and understand cutting-edge papers in related fields.

-Ability to conduct applied research in Deep Learning and Computer Vision.

-Basic knowledge in Natural Language Processing.

-Basic knowledge in Sound/Voice Recognition.

-Familiar with Python, Pytorch, Tensorflow.

-Experienced with cloud services: AWS (EC2, S3, Sagemaker, ECR), GCP.

-Experienced with Git, Linux server, CircleCI,..

-English: TOEIC with score 815.

RESEARCHES & PROJECTS

RobustOCR

Description

Our current OCR models have unstable performance on low quality, handwritten text-lines. This research is to investigate wide range of SOTA techniques to enhance current OCR library, output module can adapt different scenarios in projects with different data quality and requirements.

Technologies used

-CNN Backbones: light and heavy options.

-Sequential encoder: RNN-based, Transformer

-Decoder: Attention, CTC.

GraphKV

Description

Research and develop Graph-based models which can extract the key word and value from free-format document images.

Technologies used

- Deep learning architectures on Graph.

Table Extraction

Description

Research, benchmark and contribute to company library some solutions for table extraction from images.

Technologies used

-Object detection approach: Yolo family, RCNN based, two-stage architectures,…

-Segmentation approach: Unet and variants

-Docker, AWS SageMaker, AWS ECR,..

Skew estimation

Description

Apply traditional Computer Vision techniques to estimate skew angle of document images.

Technologies used

-Fourier Transform

-Projection-based deskew algorithm.

Vertical OCR

Description

Build a model to OCR the vertical text-line images from the original horizontal OCR engine.

Technologies used

- Linear layer, CNN

Contact this candidate