Cameron Milne
Data Scientist with experience building data-intensive applications while delivering analytical solutions to improve decision-making.
Currently at Reveal Global Consulting building language models for the U.S. Census Bureau. Previously a Data Scientist on KPMG's Data, Analytics, and AI team working on projects in the Life Sciences, Banking, and Entertainment sectors.
Portfolio
Updated: 5/31/2025
Education
- M.S. Information, University of Michigan (May 2022)
- B.A. Liberal Arts, University of Michigan (May 2020)
Certifications
- Google Professional Data Engineer (exp. June 2025)
- Google Associate Cloud Engineer (exp. March 2025)
- SnowPro Core Certification (exp. July 2025)
- Machine Learning Engineering for Production (MLOps) from DeepLearning.AI
Publications
- Ensemble Retrieval Strategies for an Improved NAICS Search Engine in the U.S. Census Bureau (Joint Statistical Meetings, 2024)
Work Experience
Sr. AI Engineer @ Reveal Global Consulting (October 2023 - Now)
- Lead developer for IRIS, an AWS-native application hosting search engines across NAICS, NAPCS, OCC, and Schedule B products, built with Docker, ECS, Fargate, and Sagemaker servicing 1000+ users internally and deployed inexpensively
- Built an LLM-driven pipeline for migrating Federal codebases in SAS to Python, indexing large codebases into graph databases and reconstructing them in Python, accepted to speak at the 2025 Joint Statistical Meetings (JSM)
- Overseeing a team of 5 data scientists on code migration for critical surveys across the U.S. Census Bureau (AHS, CPS, SOMA, NSSRN, and Frames), supporting cloud migration efforts and reducing taxpayer burden
- Designed an internal-use search engine for North American Industry Classification System (NAICS) analysts, Dockerized and wrapped in a Streamlit frontend to reduce response burden and facilitate human-in-the-loop feedback from experts
- De facto ML Engineer for the AI/ML solutions and Statistical Modernization groups through guidance on deployment best practices with CI/CD pipelines, cloud engineering, and monitoring and observability (including tools specialized for LLMs)
Data Scientist @ KPMG, Digital Lighthouse (July 2022 - August 2023)
- Designed a Google Cloud pipeline for summarizing earnings call transcripts using LangChain’s framework for chaining a VertexAI model with a VectorStore to provide contextual querying based on predefined prompts for a large banking client
- Performed attrition modeling for a leading global life sciences company, analyzing attrition for vulnerable populations with survival curves, time series analyses, and statistical inference, delivering a set of Power BI dashboards for monitoring risk
- Repaired an outdated ML pipeline predict box office results for a large streaming service using social media sentiment with 70%+ accuracy by updating deprecated API features, streamlining comment scraping, and improving regularization methods
- Presented on bootstrapping generative AI approaches for over 230 colleagues, providing tutorials on creating Streamlit.io demos with OpenAI, summarization tasks, and prompt engineering strategies for KPMG’s internal ChatGPT API with Jupyter
Natural Language Processing (NLP) Projects
Twitter Plug-In for Reducing Harmful Content
Built a Chrome extension capable of filtering a Twitter feed based on negative content (e.g. depressiveness) using an XgBoost model and packing into a RESTful API deployed on Heroku (Github, Medium)
Comparing Pre-trained and Fine-tuned Transformers Models on Patent Data
Compared performance of LLM models (BART, Pegasus, T5) before and after fine-tuning on patent data to measure generalizability of popular transformer models on scientific language and performance trade offs for size differences
Verizon Support and Sales Chatbot Service
Python implementation of a support agent capable of discerning a customer/user’s motivation and responding with the appropriate personality (tech support vs. sales) to serve their needs.
The project uses LangChain’s framework for chaining an augmented retrieval function to a chat LLM to provide document Q&A capabilities. The Agent uses the same LLM and a set of tools with predefined use-cases and retrieval mechanisms to interpret the user input and respond with the appropriate tool. The data used in this repo is a collection of Verizon FAQ materials scraped from their blog.
This is a rudimentary approach approximating a large-scale chatbot service with billing that scales with customer demand and intelligent Q&A capabilities on a personalized dataset.
Excel Document Analyzer with LangChain and VertexAI
LangChain and LLM-backed application for providing Q&A capabilities on an Excel document
SI630-NLP
My complete implementation of assignments in SI-630: Natural Language Processing at the University of Michigan (Fall, 2021)
Topics include:
- Classifying the Political Framing of Campaign Emails with PyTorch
- Training a Word2Vec model on Wikipedia Biographies and debiasing harmful associations
- Predicting Helpful Stack Overflow Answers and working with Data Annotation/Measuring Annotation Quality
- Pattern-Based Learning (Exploitation Training) for Toxic Language
Information Retrieval Projects
SI650-IR
My complete implementation of assignments in SI-650: Information Retrieval at the University of Michigan (Spring, 2022)
Topics include:
- Probabilistic models of retrieval and simple text analysis
- Building a ranking function and simple search engine
- Training a deep learning retrieval system (PyTerrier) with GPUs
Data Science Projects
Machine Learning References
Collection of ML use-cases, implementations, and cleaned datasets for reference. Scripts are sourced from personal projects and coursework.
|--PyTorch
| |-- Classifying the Political Framing of Campaign Emails (Logistic Regression)
| |-- Train a Word2Vec model on Wikipedia Biographies with debiasing (Tensorboard)
|
|--Keras
| |-- Computer Vision with CNN
|
|--HuggingFace (Transformers)
| |-- Predicting Helpful Stack Overflow Answers and Data Annotation/Measuring Annotation Quality
| |-- Pattern-Based Learning (Exploitation Training) for Toxic Language
|
|--Scikit-Learn
|
|--Coursework Examples
| |--SI630 - Natural Language Processing
| | |-- Classifying the Political Framing of Campaign Emails (Logistic Regression)
| | |-- Train a Word2Vec model on Wikipedia Biographies with debiasing (Tensorboard)
| | |-- Predicting Helpful Stack Overflow Answers and Data Annotation/Measuring Annotation Quality (HuggingFace)
| | |-- Pattern-Based Learning (Exploitation Training) for Toxic Language
| |
| |--SI670 - Applied Machine Learning
| | |--TBD
| |
| |--SI671 - Data Mining
| | |-- Mining and Evaluating Frequent Itemsets on Twitter Emojis
| | |-- Time Series analysis of COVID-19 trends for G7 Nations
| | |-- Social Network Analysis for Amazon Product Reviews
MLOps References
My complete implementation of assignments in Machine Learning Engineering for Production (MLOps) Specialization taught by Andrew Ng on Coursera. This repo is a collection of the scripts and projects for future reference.
Tech Stack
