class ColePearson:
name = "Cole E. Pearson"
age = 24
location = ["Canberra City", "Melbourne", "Australia"]
roles = [
"Lead ML Engineer",
"AI Systems Architect",
"Full-Stack ML Developer",
"Data Scientist",
"MLOps Engineer",
]
focus_areas = [
"Large Language Models & Fine-Tuning",
"Deep Learning Architecture Design",
"End-to-End ML Pipelines (Data → Production)",
"Distributed Training & Model Optimisation",
"Explainable & Ethical AI",
"Real-Time ML Inference Systems",
]
currently_exploring = [
"Agentic AI Workflows",
"Multi-Modal Foundation Models",
"RL from Human Feedback (RLHF)",
"Cloud-Native MLOps on AWS / GCP",
]
motto = "Ship intelligent systems. Explain them. Improve them. Repeat."
|
Fine-tuned a pre-trained large language model for domain-specific content generation. Implemented parameter-efficient fine-tuning (LoRA/PEFT), custom training pipelines, and automated evaluation harnesses to benchmark output quality against baseline models.
|
Production-grade NLP system for classifying and routing customer support queries using transformer-based models. Achieved significant accuracy improvements over rule-based legacy systems, deployed via REST API with real-time inference and observability dashboards.
|
|
Deep learning forecasting system using PyTorch (Temporal Fusion Transformer architecture) to predict multi-horizon sales trends and drive inventory optimisation. Integrated with data pipelines ingesting real-time POS data, with full MLflow experiment tracking.
|
Scalable REST API wrapping a fine-tuned CNN (TensorFlow / EfficientNet) for multi-class image classification. Containerised with Docker, deployed on Kubernetes with autoscaling, achieving sub-100ms p95 latency under load.
|
|
Custom RL agent using stable-baselines3 and OpenAI Gym to learn optimal navigation policies in a simulated environment. Experimented with PPO, SAC, and custom reward shaping. Visualised learned policies and training curves with Weights & Biases.
|
End-to-end ML pipeline for a telco churn prediction use case — from EDA and feature engineering to model selection (XGBoost, Random Forest, logistic regression ensemble), SHAP-based explainability reporting, and a Streamlit monitoring dashboard.
|
██████╗ ██████╗ ██████╗ ██████╗ ██████╗ ██████╗ ██████╗ ██████╗
╚════██╗██╔═████╗╚════██╗██╔════╝ ██╔════╝ ██╔═══██╗██╔════╝ ██╔════╝
█████╔╝██║██╔██║ █████╔╝███████╗ ██║ ███╗██║ ██║███████╗ ███████╗
██╔═══╝ ████╔╝██║██╔═══╝ ██╔═══██╗ ██║ ██║██║ ██║██╔═══██╗╚════██║
███████╗╚██████╔╝███████╗╚██████╔╝ ╚██████╔╝╚██████╔╝╚██████╔╝██████╔╝
╚══════╝ ╚═════╝ ╚══════╝ ╚═════╝ ╚═════╝ ╚═════╝ ╚═════╝ ╚═════╝
| Status | Goal |
|---|---|
| ✅ | Ship a production ML feature end-to-end — own the full lifecycle: ideation, training, deployment, monitoring |
| ✅ | Contribute to an open-source ML library — merged PRs in a recognised OSS project |
| ✅ | Publish a technical deep-dive — blog post or paper on model optimisation, LLM fine-tuning, or MLOps patterns |
| ✅ | Earn AWS ML Specialty or GCP Professional ML Engineer certification |
| 🔲 | Build and deploy an autonomous AI agent — multi-step reasoning, tool use, real-world task completion |
| 🔲 | Lead a data team initiative at scale — mentor junior engineers, define ML standards and best practices |
| 🔲 | Architect a real-time ML inference system — streaming data, sub-50ms latency SLA, full observability |
| ✅ | Design a full-stack AI application — from data ingestion layer to React/Next.js frontend, prod-deployed |
| ✅ | Publish open-source ML tooling — a reusable library or template repo that others actually use |
| 🔲 | Master distributed training — multi-GPU / multi-node training on large models with DeepSpeed or FSDP |
"The best model is the one in production — properly monitored, continuously improving, and explainable to the people it affects."