Banking System - US
AI Developer, Aug 2024 - Present
Project description
- A platform that enables the abilities to gather data from multiple sources, process and transform it into usable formats and store it in a centralized feature store to support building ML models for analysing business trends.
- Including automation pipelines and jobs for batching, scheduled or streaming data processing.
- Empower the GenAI capabilities by hosted an internal RAG system with agent mechanism, including MLFlow, Langchain & Vector database.
Responsibilities
- Architected and developed features within OCB Data Platform, and Data/ML platform that serves multiple internal ML models with Databricks on AWS.
- Developed and deployed a GenAI RAG system on Databricks for internal usage, using Langchain, MLflow and indexing vector database.
- Managed and developed AWS and Databricks infrastructure with multiple pipelines and plans with Terraform and Jenkins
- Developed data pipeline including ETL/ELT, feature engineering workflows with AWS DMS for cutoff and CDC plans, ... with Spark and Airflow as automation jobs.
- Implemented MLOps workflow with Databirck Assert Bundle (DAB) including CI/CD pipeline to build, test, and deploy code to production quickly and continuously through environments.
Technologies
- AWS, Databricks, Apache Spark, Terraform, Jenkins, Python, Airflow, Oracle, MySQL, Networking.
Telecom - Hongkong
AI Developer, March 2022 – Aug 2024
Project description
- Tessel AI is a full-stack multi-model AI platform designed for retail companies to manage and evaluate their in-store advertising campaigns using computer vision.
- The platform uses AI models to verify images captured from stores, ensuring ads are correctly displayed and compliant with brand guidelines.
- The platform includes the entire machine learning lifecycle, from automated data labelling to user-defined model training and deployment.
Responsibilities
- Architected and developed Tessel AI project, an AI platform where users can label, train, and deploy ML models.
- Optimized model’s inference latency from 2 seconds to 350 milliseconds per image by batching requests.
- Built Human-in-the-loop workflow, effectively collects predicted data on production to retrain models.
- Developed Auto-label workflow by using a foundation model to reduce labelling time from 1 minute to 15 seconds.
- Implemented CI/CD pipeline to build, test, and deploy to production quickly and continuously
Technologies
- Python, Pytorch, Triton, AWS, Fast API, Flask, PostgreSQL, Docker, Kubernetes, Redis.
Computer vision system - USA
AI Developer, Jun 2021 – March 2022
Project description
- A computer vision system to help car insurance companies automatically evaluate vehicle damage for claims processing.
- Leveraged object detection and instance segmentation models to identify and analyse damaged car parts from user-submitted images.
- Including data ingestion pipeline using Airflow to feed labelled data into the training workflow and MLflow for experiment tracking and model improvement.
Responsibilities
- Built software to extract insurance data from vehicle images, achieving an accuracy up to 90%.
- Improve performance of YOLO and Detection model by multi-stages inference.
- Significantly boosted the accuracy of instance segmentation system (62 to 85%) by post processing the segmentation output as contours.
- Deployed backend/ infrastructure to the AWS cloud environment.
Technologies
- Python, AWS, Pytorch, OpenCV
Data platform
AI Developer, Jan 2021 – Jun 2021
Project description
- A system to collect and analyse text data from online forums, transforming raw discussions into structured insights.
- Extracted user feedback and trending topics to support data-driven business decision-making.
Responsibilities
- Developed system crawls text from e-forums and extract data as feedback and trending for making business decisions.
- Implement crawler using selenium and beautiful soup.
- Develop a BERT classifier for sentiment analysis
Technologies
- Python, Git, Transformers, BERT, Flask
