Canada

AI/ML Engineering

Daryl Fung

Machine Learning Engineer

Problem-solving

Time management

AI/ML

Python

PyTorch

React

React Native

Jupyter Notebook

Redis

- Conducted extensive research in computational science modeling, with a specialization in the development of deep learning algorithms. Particularly focused on leveraging generative AI techniques to solve engineering physics-driven problems such as flows and acoustics.
- Led AI developer in the design and implementation of advanced AI models for forecasting of pressure cavities, enhancing predictive accuracy and optimizing experimental outcomes.
- Collaborated with professor and researchers to publish research papers in high-quality journals, demonstrated strong communication and teamwork skills throughout the publication process.
- Created compelling data visualizations to effectively communicate complex findings to both technical and non-technical stakeholders. Employed visualization tools and techniques to enhance understanding and facilitate informed decision-making.
- Demonstrated a strong ability to tackle complex problems by applying innovative approaches and leveraging cutting-edge methodologies. Utilized deep learning algorithms to optimize solutions and drive impactful results.
- Carried tests to select the best performing deep learning model for interpolating fluid motions.

- Successfully managed 3 ongoing projects simultaneously, by planning, organizing, and directing projects across Canada within Health Research Network (HDRN) Canada to completion, successes in projects delivered increases data category labels by 30%.
- Collaborated with data analyst, researchers, and working groups across Canada within Health Data Research Network (HDRN) Canada to coordinate work, develop project goals and organize available resources.
- Coordinated project tasks and act as a point of contact for project collaborators, ensuring effective communication and timely completion of project milestones.
- Implemented and applied natural language processing models including key phrase extraction (TextRank), keyword extraction (KeyBERT), named entity recognition (Spacy) to extract key information, enabling more informative descriptions for data analysis
- Created a knowledge graph using neo4j that provides clear visualisation of dataset across Canada within Health Data Research Network (HDRN) Canada to enable increased efficiency for researchers to access datasets located in Canada
- Developed automated data pipeline with ClearML to facilitate the flow of data while eliminating the need for data cleaning manually.

- Planned and organised comprehensive training programs for players tailored to their skill level to ensure the most efficient development of their performance
- Evaluated players' performance during drills and practices and provided personalised feedbacks to help players reach their full potential
- Ensured players' safety by enforcing the proper usage of equipments while maintaining a safe and clean environment for practice

- Managed technology stack for CEO and CRO and communicated with front-end engineer so that the software architecture was up to date for the demonstration to investors
- Launched and maintained machine learning toolkit to conduct analysis on language (NLP) and audio (speech processing) as microservices with Docker for better scalability which enabled the accessibility to demonstrate for investors.
- Created continuous integration and continuous deployment (CI/CD) pipelines for the server to ensure code consistency and maintainability while preventing new code to break old code
- Stored raw and preprocessed customers' NLP features into a PostgreSQL database for easy access through an API for both customers and engineers
- Documented machine learning toolkit thoroughly and concisely which helps onboard engineers 50% more efficiently
- Generated word cloud bubbles to enable clear data visualization on the importance of words to present to investors which increases chances of generating investment for the company
- Generated knowledge graph and a variety of word embeddings visualization to analyze texts and determine relationships between words which helps improved context comprehension of documents
- Discussed with co-founders the development of the server with new features to achieve investors' requests effectively

- Deployed an end to end machine learning pipeline that includes data cleaning, data preprocessing, model prototyping, model training, model evaluation, and production deployment with experimental tracking.

- Developed backend server in AWS(Elastic BeanStalk) for the hosting of deep learning models to receive still images from clients to generate an animation of the still images to be sent back to client


http://darylfung96.1p21.io:5004/
- Implemented front-end user interface to receive images and audio to be sent to the backend to generate a face and head animation 
- Created deep learning algorithms using a first-order motion model combined with Wav2Lip to generate a talking head with a single audio file and a single image for the client
- Created Backend API using FastAPI to receive requests from customers to generate a talking head for the customer to achieve business requirements

- Improved reports of heart ventricle segmentation using deep learning by providing suggestions and feedback as needed to resolve deficiency of reports which resulted in FDA approval

- Led and managed building machine learning tools for music generation till the success of publishing into production
- Communicated effectively with business clients to determine the best technologies to realize their goals and visions
- Improved model speed by 43% using model optimization technique with pruning, quantization, and model compilation in PyTorch.
- Integrated computer vision and deep learning techniques to create animation for face images through head guidance from a moving head in a video
- Incorporated generation of music using GAN  and MuseGAN that creates music from scratch or conditioned on a track by hosting in DigitalOcean with Docker
- Stored user information in SQL so that users can easily retrieve songs generated by GAN through restAPI in a structured way
- Developed load balancing service to dissociate requests to multiple servers which results in a speed up of 50%.

- Improved deep learning algorithms by 0.17 AUC to analyze the longitudinal study of the gut microbiome to classify disease severity and remission of ulcerative pediatric patients
- Extended LassoNet and Autoencoder with batch loss and robust loss as a one-step approach to incorporate batch effects from several studies as an end-to-end network with an improvement of 0.14 AUC to baseline models
- Integrated a self-supervised computer vision (CV) deep learning model to undergo infected region segmentation of CT lung images of patients infected with COVID-19 using InfNet as one of the baseline model which improved F1 score by an average of 0.11 
- Utilised and fine-tuned a computer vision (CV) state-of-the-art object detection (YOLOv5l6) algorithm on finger/wrist joints to be able to detect joints of rheumatoid arthritis patients
- Communicated the algorithm created effectively for my peers and supervisors to aid in the understanding of the workings of the algorithm

- Prepared and planned materials to explain concepts and theories in an easy and understanding manner
- Elaborated study materials and broken them down into hierarchical sections and diagrams depending on student's best way of learning style
- Communicated effectively and helped students prepare for their exams so that they are well-prepared while being calm and relaxed

- Implemented a deep learning algorithm that is able to recognise a different view of the actions when trained on a different angle by using maximum mean discrepancy to minimise the domain differences.
- Discussed with professor to find the different datasets to compare on with the other baselines of action recognition
- Utilises CUDA to speed up the training for the cross view action recognition

- Utilized DialogFlow for the implementation of a chatbot to understand customers' responses using natural language processing (NLP) while guiding customers' conversations to the appropriate direction.
- Discussed and collaborate with clients to prioritize features to be implemented in the chatbot.
- Implemented an email service that sends a report when a customer finished chatting with the chatbot to keep a history of data to be used for future training.

- Integrated motions and computer visions (OpenCV) into a robot for a variety of events held at the FIRA RoboCup 2017 competition in Taiwan.
- Implemented object detection through computer vision for various events.
- Carried thorough tests to ensure stability of robot in preparation for competition.
- Collaborated with teammates on robots’ motions by determining, analyzing solutions and solving the problems appropriately.

- Formed a MapleStory game server with easier access to environment and obtaining items with less difficulty with C++.
- Designed a dynamic link library (DLL) in C++ that gets injected into MapleStory game server to enhance and modify game behaviour.

Daryl Fung🇨🇦

Create an account to accelerate your career with expert mentors worldwide.