ML Engineer — Automated Property Valuation

Chloe Martin

Freelance · Mid-level · London

This role is no longer accepting applications

Similar open roles you can apply to now:

About the role

We build automated property valuation software for mortgage lenders, conveyancers, and property investors in the UK market. Our Automated Valuation Model produces indicative valuations for residential properties at scale — we process approximately 85,000 valuation requests per month on behalf of our lending clients. The model that underpins those valuations was built three years ago. It is based on a gradient boosting approach trained on Land Registry transaction data, EPC certificates, and our own feature engineering on property attributes. It performs well on standard residential stock in areas with high transaction volume. It performs poorly on unusual properties, properties in areas with sparse comparables, and properties where the attributes in the public record are inaccurate. We know this. We are commissioning an eight-week project to evaluate whether ML approaches — including neural architectures and alternative feature engineering strategies — can improve performance on these tail cases without degrading performance on the core residential stock where the existing model already works. This is a research and evaluation project, not a production engineering engagement. We want rigorous analysis, clear documentation, and an honest assessment of what is worth building next.

Key Deliverables

List the expected deliverables for this project.

Audit the performance of the existing AVM model by property type, geography, and comparables density — producing a clear diagnostic of where performance gaps are largest
Design and evaluate alternative ML approaches for the tail cases identified in the audit
Build a feature importance and error analysis framework that the internal data team can use for ongoing model monitoring
Produce a written technical report with evaluation results, honest assessment of what works and what doesn't, and recommendations for the next development phase
Conduct a structured handover session with the internal data team

Requirements

Technical stack needed for this mission.

4+ years of ML engineering or data science with experience in property valuation, real estate analytics, or structured tabular prediction problems
Python for data processing, model development, and building reproducible evaluation pipelines
Scikit-learn for gradient boosting, feature importance analysis, and the classical approaches that remain competitive on structured tabular data
NumPy for numerical feature engineering and the array manipulation required by custom model components
SQL for working with Land Registry transaction data, EPC data, and our internal valuation request logs
Statistics: understanding model uncertainty quantification, the specific challenges of small-sample evaluation on rare property types, and the right way to report model performance when the distribution is heterogeneous
Familiarity with UK residential property market data — Land Registry, Rightmove/Zoopla attributes, EPC certificates, council tax bands — is a significant practical advantage