Sophie P.

Data scientist with 4+ years of experience, specialising in LLM and NLP systems that run in production and are measured against real benchmarks. At a global insurance tech firm, I built a RAG system on AWS Bedrock that replaced an external AI vendor. It improved accuracy by 8% and reduced projected vendor costs by £50k+ a year. Recent results: - Delivered 7 BERT classification and NER models 3 months ahead of a 6-month deadline, meeting regulatory requirements and improving claim handling efficiency by 44%. - Improved address matching accuracy from 81% to 97%, supporting a £5m public health research partnership. - Built a live air quality dashboard for a local authority Chief Executive’s Office, refreshing automatically from a public API. How I can help: - LLM and RAG systems: classification, extraction and Q&A over your own documents, with clear evaluation so you know how well they perform. - Custom NLP models: fine-tuned transformers deployed with Docker and AWS SageMaker. - Data pipelines and dashboards: SQL, dbt, Snowflake, Power BI, Streamlit. I also teach Python and SQL with Code First Girls, where I’ve worked with 300+ students, so I’m comfortable explaining technical work in plain terms. Stack: Python, SQL, PyTorch, HuggingFace, AWS (Bedrock, SageMaker), dbt, Snowflake, Docker, Power BI. If you have a project in mind, I’d be happy to talk through how I’d approach it.