Skip to content
View Stephaniew1's full-sized avatar

Achievements

Achievement: QuickdrawAchievement: Pull Shark

Achievements

Achievement: QuickdrawAchievement: Pull Shark

Block or report Stephaniew1

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Stephaniew1/README.md

Hi there, I'm Stephanie 👋

I'm a Machine Learning Engineer based in Melbourne, currently leading research into the productionisation of an agentic chatbot at FocusBear. I designed and built an agentic AI assistant end to end: from the architecture research the team adopted, through retrieval and evaluation, to live integration with the production backend and the AWS infrastructure it runs on. I hold a Bachelor of Applied Data Science from Monash University.

I care about building ML systems that solve real problems for real users, with a strong preference for code-owned, transparent architectures over black-box platforms.

What I work on

  • LLM applications and agents: retrieval-augmented generation, function-calling agents, prompt engineering, RAG evaluation
  • Production ML systems: serverless deployment, infrastructure as code, authentication, testing, architecture documentation
  • Healthcare and scientific computing: ICU mortality prediction, molecular image analysis, climate driver modelling

Tech stack

Languages: Python, TypeScript, SQL, R LLM and ML: LlamaIndex, LangChain, LangGraph, OpenAI, HuggingFace Transformers, PyTorch, scikit-learn, RAGAS Retrieval: pgvector, FAISS Backend and cloud: FastAPI, AWS (Lambda, CDK, API Gateway, Secrets Manager, CloudWatch), Docker, PostgreSQL, JWT and OAuth 2.0 Data and viz: pandas, NumPy, Plotly, Tableau, Power BI Tooling: Git, pytest, LocalStack, Streamlit, Google Cloud

Featured projects

Agentic RAG Assistant

Production-shaped AI assistant that answers questions from a knowledge base with cited sources and acts on a user's account through function-calling tools. Features per-user JWT authentication, prompt-injection defence for uploaded documents, lazy serverless initialisation, and an AWS CDK stack with an API Gateway JWT authorizer. A clean-room reimplementation of patterns from my FocusBear work, with a case study covering the decisions behind it.

Invisible Disability Assistant: Nairobi Transport AI

Bilingual English and Swahili RAG assistant for navigating Nairobi's Matatu network. Built with FAISS, HuggingFace, ChatGroq and deep-translator, with a conceptual offline GPT-2 extension for low-connectivity areas.

Deep Learning for Sequential Data

PyTorch implementations of RNN, GRU, LSTM and BERT for NLP question classification. Includes a from-scratch RNN built on raw tensor operations, a configurable BaseRNN with multiple pooling strategies, and fine-tuned BERT on the TREC dataset.

ICU Patient Mortality Predictor

Model predicting ICU mortality risk from the five most clinically critical vital signs. Achieved ROC-AUC of 0.75 to 0.82, with an emphasis on interpretability for clinical decision support.

Counting Molecules

Computer vision pipeline for automated counting and classification of molecule species from STM imaging data. Combined binarisation, blob detection and scikit-learn classifiers to reduce manual inspection effort by 40 to 50 percent.

Modelling Climate Drivers

Descriptive analysis and modelling of climate variables, developed during my CSIRO Aspendale placement, where I analysed post-wildfire atmospheric data and built dashboards and reports that informed research planning.

Connect

Open to Data Science, Machine Learning Engineering, Forward deployed Engineer, AI Engineering and Data Engineering roles.

Pinned Loading

  1. Birthday-Invite Birthday-Invite Public

    interactive web-based party invite app with chatbot-style RSVPs and automated calendar invites. Built for a surprise birthday party before Apple launched Apple Invites

    Python

  2. Counting-Molecules Counting-Molecules Public

    Computer vision pipeline that automates molecule counting and classification from Scanning Tunneling Microscope (STM) imaging data. Built with scikit-image, OpenCV, and scikit-learn for a Monash da…

    Jupyter Notebook

  3. deep-learning-sequential-data deep-learning-sequential-data Public

    PyTorch implementations of RNN, GRU, LSTM, and BERT-based models for NLP question classification. Includes a from-scratch RNN forward pass, configurable BaseRNN with pooling strategies, and fine-tu…

    HTML

  4. ICU-project ICU-project Public

    Develop a model to predict patient mortality rate by analyzing the five most important vital signs of ICU patients

    Jupyter Notebook

  5. Stock-Portfolio-Optimization Stock-Portfolio-Optimization Public

    In this project, we perform an in-sample optimization of trading portfolios, based on the stocks which have been in the S and P 500 in the last 16 years.

    Jupyter Notebook

  6. Agentic-rag-assistant Agentic-rag-assistant Public

    Production-shaped AI assistant: RAG over a knowledge base plus agentic tools that act on a user's account. FastAPI, LlamaIndex, pgvector, per-user JWT auth, prompt-injection defence, AWS CDK (Lambd…

    Python 1