Peikai Tan

- Led and scaled an 8-member, cross-functional team to design and deliver cloud-native MLOps pipelines enabling reliable retraining and deployment of production deep learning models - Architected a model data engine to automate performance monitoring, evaluation, and continuous improvement across multiple deep learning systems - Developed a proof of concept for confidence-based performance estimation of production models, providing actionable insights with minimal latency, compute overhead, and data requirements - Mentored and coached high-potential interns through hands-on technical leadership, resulting in successful full-time conversions and placements at top industry organizations - Owned end-to-end project planning and execution, including stakeholder alignment, delivery timelines, and change management across engineering and business teams

- Led and scaled a globally distributed data engineering team to deliver cloud-native, event-driven pipelines supporting large-scale document intelligence and information extraction - Architected and spearheaded the end-to-end development of a document synthesis platform from the ground up, significantly improving model accuracy while reducing dependency on large training datasets - Designed and built a full-stack data observability and governance application, enabling improved visibility, trust, and self-service access for stakeholders across the data lake - Leveraged large language models (LLMs) for synthetic data generation, augmenting low-resource datasets and expanding model coverage across edge cases - Architected and implemented a serverless, real-time data flywheel enabling continuous human-in-the-loop feedback to iteratively improve multiple production deep learning models - Implemented scalable data lake clustering and storage strategies to efficiently manage and retrieve multi-format data across millions of records

- Built, deployed, and maintained a scalable data pre-annotation pipeline using Docker, AWS Lambda, and SQS, increasing data annotation efficiency by 200% - Improved scheduled model retraining outcomes by prioritizing annotation workloads based on machine-generated confidence scores - Reduced cloud infrastructure costs by 300% through algorithm optimizations and efficient EC2 resource tuning - Collaborated closely with machine learning engineers and annotation teams to streamline data feedback loops, improving model performance while reducing manual effort - Implemented monitoring and alerting for data pipelines using AWS CloudWatch and custom health checks, cutting debugging and incident response time by 60%

As the sole Machine Learning Engineer supporting a team of Data Scientists, I worked closely with my team to: - Develop a custom signal processing audio denoising algorithm that outperforms deep learning model and improved subsequent deep learning model performance by 7%. - Pioneer the use of pretrained audio neural network embeddings to guide a diffusion deep learning model to improve audio data quality, allowing 20% more of existing dataset to be used for training - Create a synthetic audio dataset using generative models and audio processing techniques - Create a high quality synthetic image dataset using blender for shadow generation and reflection removal - Architect, develop and maintain an automated cloud distributed AI training framework utilising Terraform, Bash and Python to reduce operational cost and manhours - Lead a team for the development of a recommendation system solution involving GPT-4, Pinecone, Next.js and Flask - Implement a quality assurance and performance monitoring pipeline using Python, Docker, Terraform, AWS Glue, AWS Athena, Fargate and Tableau, improving efficiency by 300%

As a Software Developer at JTC Corporation, duties and responsibilities include: - Architecting an in-house frontend framework on Unity 3D for JTC's Open Digital Platform. - Create, test, and develop components for the framework. - Porting of existing Open Digital Platform projects to the in-house framework.

Research work on a Mixed Reality Project involving Microsoft’s HoloLens 2 for Design and FEA Analysis. - Create, test, and develop Mixed Reality application on Unity 3D for the HoloLens 2. - Develop, optimise algorithms to increase performance of application - Research and implements ways of visualizing FEA data in Mixed Reality - Communicate and work closely with engineers and clients to deliver product.

- Oversaw and facilitated day to day training exercises and activities with the platoon - Conducted and provided safety for training activities - Successfully participated in major exercises with my company