Australia

Scott Yip

Data Scientist & Software Engineer building production GenAI and ML systems

7M: App customers reached
3: Production systems shipped
Dux: Master of Data Science
3: Cities worked in; Melbourne · Auckland · SF

From balance sheets to machine learning

In 2019, after a long day of spreadsheets, I came across a Medium article on neural networks. After staring at it for far too long, dazzled and confused, I decided to retrain in data science. I went on to graduate dux of my Monash Master's, and now build production GenAI and machine learning systems at Commonwealth Bank.

My consulting background still shapes how I work: I combine commercial acumen and a feel for what actually moves the business with hands-on technical depth, working in the trenches to build the systems.

What I enjoy most is owning problems end to end, from the model to the product people actually use. Away from the screen, I'm usually coding side projects, training at the gym, travelling, getting far too into interior design, or out with my camera.

Highlights

Built CBA's first customer-facing GenAI agent: 7M customers · 200+ disputes a day
Unblocked voice agents with novel guardrails: 0.997 AUROC · latency 1,800ms to 49ms
Forecast cashflow for business banking customers: 2M customers · time-series ML in production
Shipped a frauds & scams agentic workflow: ~30 min saved per case, for 200 analysts
2025 CBA Excellence Awards finalist: for the customer-facing Disputes Agent

Toolbox

GenAI & Agents

LLMs
Agentic workflows
LangGraph
ADK
PydanticAI
Guardrails
LLM evaluation
Fine-tuning

Machine Learning

NLP
Classification
Forecasting
Fraud / risk
Embeddings
pgvector

Backend & Data

FastAPI
PostgreSQL
Redis
SQLAlchemy
SQL
Apache Spark
PySpark
Spark SQL
Database design

Infra & Observability

Docker
AWS
SageMaker
HyperPod
Linux
Vercel
Railway
Logfire

Languages

Python
SQL
R

Built in production

2023 - Present Now

Data Scientist, GenAI

Commonwealth Bank of Australia · Melbourne, VIC

Building production GenAI systems across the bank: agents, guardrails, and fraud/scams workflows.

Built CBA's first customer-facing GenAI agent (disputes): live to 7M app customers, serving 200+ disputes/day, and a 2025 CBA Excellence Awards finalist.
Cut dispute rework 12% with a structured LangGraph agent workflow; 96% intent-classification and 94% information-extraction accuracy.
Shipped a frauds & scams agentic workflow used by 200+ analysts, synthesising customer, crypto, fraud-check and ML signals, saving ~30 min per case.
Fine-tuned streaming financial-advice guardrails, cutting latency from ~1,800ms to 49ms at 0.88 F1 / 0.997 AUROC.
Nov 2021 - Mar 2022

Summer AI Research Assistant

Monash University · Melbourne, VIC

Transformer-based NLP research.

Prototyped a Transformer model for multi-label text classification, benchmarking encoder-decoder and BERT-based architectures across applied NLP tasks.
Mar 2016 - Mar 2021

Audit, Taxation & Advisory Manager

Pitcher Partners · Melbourne · Auckland · San Francisco

Managed audit, tax and advisory engagements (~AUD 450K p.a.), blending accounting with data-driven analysis.

Built a firm-wide AASB 16 leasing model adopted across a 400-person division.
Implemented a bank-transaction classification algorithm for forecasting, saving 40+ hours of manual work.
Revamped graduate training onto Rise360, contributing to a 6% lift in training satisfaction.

Projects

Things I've built

Culina

AI-Powered Nutrition & Macro Tracking App

Built and launched a production AI nutrition app with React/Vite, FastAPI, PostgreSQL, Redis and SQLAlchemy, with LLM-based food entry generated from text and photos.
Implemented semantic food search with pgvector nearest-neighbour retrieval, image-derived embeddings, Logfire observability, rate limiting and optimised SQL macro aggregation.