Trenton Potgieter¶
Principal AI Engineer
LLMs & Generative AI | Cloud Architecture | MLOps & LLMOps | AI in Gaming
View Publications View Projects View Community Enablement Get in Touch
About Me¶
An accomplished AI/ML expert with extensive experience developing and deploying large language models, machine learning systems, and distributed AI applications in production. I leverage deep technical expertise in cloud computing, ML frameworks, and distributed computing to design scalable, high-performance solutions.
With a proven track record of leading cross-functional global teams, I drive innovation and deliver impactful thought leadership through public speaking engagements, publications, technical blogs, and open-source contributions.
AI & Machine Learning¶
- Classical Machine Learning (Vision & NLP)
- Large Language Models (LLMs)
- Generative AI & Agentic Systems
- Supervised Fine-tuning & Alignment Techniques
- Model Evaluation & Benchmarking
- Data Lakes, Graph & Vector Databases
AI Cloud & Infrastructure¶
- AWS Solutions Architect
- AWS Machine Learning Specialist
- Kubernetes & Docker
- Serverless & Event-driven Architecture
- CDK, Terraform & Infrastructure as Code
MLOps & LLMOps¶
- SageMaker & MLflow
- Model Serving & Inference
- CI/CD for ML Pipelines
- Experiment Management
- Multi-tenant Production ML Systems
Publications¶
Book Automated Machine Learning on AWS
The official guide highlighting the MLOps process on AWS. Covers end-to-end machine learning automation, from data preparation to model deployment and monitoring.
Book (Co-Author) Applied Machine Learning and High-Performance Computing on AWS
Comprehensive guide on how to build, train, and deploy large machine learning models in production, and at scale on AWS, using high-performance computing techniques.
Key Projects¶
Project Ownership¶
Agentic AI Dialog Engine¶
A cloud-native, GitOps-driven platform delivering low-latency generative AI-powered NPC characters with distinct personalities, behavioral consistency, and emotional authenticity at scale across multi-tenant game environments.
Personality Fine-Tuning Research¶
Architected comprehensive experiment pipeline evaluating multiple model variants (evolutionary alignment, PEFT, model merging configurations), proving smaller specialized models achieve parity with larger base models for character-specific personality, and behaviors.
Automated Training Data Distillation¶
Engineered automated distillation infrastructure transforming teacher model outputs into custom training datasets, with sufficient character-consistent tokens to modify parameters across large as-well-as small parameter models.
Dynamic NPC Dialogue on AWS¶
Architected AWS reference solution enabling dynamic NPC dialogue with RAG-powered knowledge retrieval and personality-driven responses using Amazon Bedrock. Includes LLMOps pipelines, fine-tuning workflows, and vectorized game lore integration.
Predicting Player Behavior with AI on AWS¶
Architected automated ML pipeline enabling game studios to predict player retention, engagement, and monetization in real-time. Leverages SageMaker Autopilot for end-to-end model lifecycle management—from data ingestion through deployment—without requiring manual ML expertise.
Project Leadership¶
Game Analytics Pipeline on AWS¶
Led the development of a modular, serverless analytics pipeline helping game developers transform raw gameplay data into actionable insights. Supports multi-game, multi-tenant, and multi-region deployment with Data Lake or Data Warehouse configurations for real-time and batch telemetry processing.
Multi-Agent Builder with Bedrock & CrewAI¶
Conceived and lead the no-code platform for building AI agent teams powered by Amazon Bedrock and CrewAI. Contributed the initial backend prototype, enabling orchestration of specialized agents for complex multi-step tasks across content creation, code generation, and visual design workflows.
Project Contributions¶
Game Tech Cohort Modeler on AWS¶
Contributed to graph-native solution enabling game studios to map player relationships and classify behavioral patterns within their player base. Leverages Amazon Neptune for relationship discovery, collaborative filtering, triadic closure, and bad actor detection through serverless prediction APIs.
Custom Game Backend Hosting on AWS¶
Built Delta Lake integration feature enabling game developers to connect backend telemetry to Databricks lakehouse architecture for large-scale analytics. Contributed to multi-platform authentication framework supporting Unreal, Unity, and Godot with cross-platform identity management, enabling the Databricks partnership and analytics blog publication.
Enablement¶
Workshops¶
Upgrade Your Game Storyboards with Multimodal Prompt Chaining¶
Built and delivered a hands-on workshop on multi-modal, agentic AI systems for game narrative development.
Operationalize Generative AI Applications Using LLMOps¶
Built and delivered the comprehensive LLMOps workshop covering best practices for deploying and managing generative AI applications in production.
Amazon SageMaker MLOps: From Idea to Production¶
Built and delivered a comprehensive MLOps workshop covering the complete ML lifecycle—from experimentation to production deployment using SageMaker Pipelines, CI/CD automation, model monitoring, and drift detection.
Blogs¶
Operationalize Generative AI: Part I - LLMOps Framework Overview¶
Foundational overview of implementing Large Language Model Operations (LLMOps) on AWS. Covers the comprehensive framework for managing generative AI applications in production, including model lifecycle management, monitoring strategies, and the essential AWS service ecosystem.
Operationalize Generative AI: Part II - Production Architecture Deep Dive¶
Technical architecture deep-dive into building production-ready generative AI systems. Explores scalable design patterns, CI/CD automation with CodePipeline, service integration strategies, and best practices for deploying LLM applications using Amazon Bedrock, SageMaker, OpenSearch, and Lambda.
Maximize Your Game Data Insights with the Updated Game Analytics Pipeline¶
Deep-dive into the modernized Game Analytics Pipeline solution, now built with AWS CDK for modular deployment. Covers the DataOps CI/CD pipeline, latest AWS Glue optimizations, and how to integrate AI/ML capabilities for fraud detection, player promotions, LiveOps automation, and real-time insights.