Machine learning & NLP
BERT and BiLSTM classifiers, a BERT-based recommendation model, and leakage-aware evaluation design for real-world prediction problems.
Who I am
Priscilla Miller
I’m a former editor turned Berkeley-trained data scientist seeking a data science, AI engineering, or ML role.
What interests me
Clean prose, clean code, all things data
For over a decade at KPMG, I was an editor ensuring documents were clear, accurate, and compliant. But I kept gravitating toward the technical side, even building the occasional automation script to help those around me work smarter and more efficiently. I wanted to keep learning and growing, and I was ready to move on from editing what others wrote to building things myself.
So I went back to school. UC Berkeley's Master of Information and Data Science program taught me to build ML models, work with messy data, and deploy systems that actually solve problems.
Now I'm looking for roles that combine my editorial experience with my technical skills, particularly in data science, AI engineering, or ML-focused work at a company that values transparency and collaboration.
What I do
Clarity from chaos
Data analysis
SQL analysis from raw public records to a published dashboard: loading and cleaning in PostgreSQL, window functions and CTEs for trends, and charts that update when the analysis is rerun.
AI-native engineering
Apps, tools, and devices built on Claude: RAG chatbots, LangGraph pipelines with bounded revision loops, and node contracts that stop vision models from inventing label text.
What I have done
Featured work
Conversational AI · hardware
Purrserpina
A talking Halloween cat on a Raspberry Pi. She wakes when she sees a person, listens with faster-whisper, replies in character through Claude, and moves her jaw in time with Piper speech.
What’s on my desk
In progress...
- Data scienceFreelance data scientist assessing a chargeback-prediction model for a subscription-billing company: label definitions, target leakage, and calibration across segments.
- Model evaluationAI model evaluation specialist at DataAnnotation and Handshake AI, judging model responses for accuracy, reasoning quality, and instruction adherence.
- StudyingLinguistics, Stanford’s CS224N, and post-training methods such as DPO and RLHF.
- Based inNapa Valley, California. Remote first; open to hybrid in San Francisco or Sacramento.