Implementing AI in a UK Property Data Platform
Challenge
The client runs one of the UK’s most popular PropTech SaaS platforms for estate agents. Its USP is an unrivalled property database used by tier-one brands such as Knight Frank and Carter Jonas. But competition is heating up and users now expect AI-powered insights in near-real time. Existing machine-learning models were slow, sparsely documented and hard to iterate on. New data pipelines had to be designed around confidential on-prem datasets, ruling out the public cloud. The internal data-science team lacked bandwidth to rebuild the infrastructure before the next big product launch, and any delay risked losing market share to fast-moving rivals.
The client runs one of the UK’s most popular PropTech SaaS platforms for estate agents. Its USP is an unrivalled property database used by tier-one brands such as Knight Frank and Carter Jonas. But competition is heating up and users now expect AI-powered insights in near-real time. Existing machine-learning models were slow, sparsely documented and hard to iterate on. New data pipelines had to be designed around confidential on-prem datasets, ruling out the public cloud. The internal data-science team lacked bandwidth to rebuild the infrastructure before the next big product launch, and any delay risked losing market share to fast-moving rivals.
Solution
Sigli embedded a data engineer and a data scientist into the client’s scrum.
1. Architecture & ML audit – we reverse-engineered legacy Python notebooks, documented data flows and benchmarked model latency.
2. Data-pipeline engineering – using Apache Airflow and Pandas we built dozens of reusable ETL pipelines that ingest land-registry, EPC and third-party lead-gen feeds into a unified Parquet lake.
3. ML modernisation – legacy XGBoost models were re-trained in scikit-learn/TensorFlow, containerised with Docker and orchestrated via on-prem Kubernetes.
4. DevEx & CI/CD – GitLab CI, unit-tests and auto-linting reduced merge friction; a custom metadata tracker stored experiment metrics for faster iteration.
Throughout the engagement we paired daily with the in-house team, authored full run-books and transferred knowledge to make sure the client would stay self-sufficient after hand-off
Sigli embedded a data engineer and a data scientist into the client’s scrum.
1. Architecture & ML audit – we reverse-engineered legacy Python notebooks, documented data flows and benchmarked model latency.
2. Data-pipeline engineering – using Apache Airflow and Pandas we built dozens of reusable ETL pipelines that ingest land-registry, EPC and third-party lead-gen feeds into a unified Parquet lake.
3. ML modernisation – legacy XGBoost models were re-trained in scikit-learn/TensorFlow, containerised with Docker and orchestrated via on-prem Kubernetes.
4. DevEx & CI/CD – GitLab CI, unit-tests and auto-linting reduced merge friction; a custom metadata tracker stored experiment metrics for faster iteration.
Throughout the engagement we paired daily with the in-house team, authored full run-books and transferred knowledge to make sure the client would stay self-sufficient after hand-off
Results
30+ production-grade data pipelines now refresh proprietary datasets hourly, enabling new prospecting features for agents.
Up-to-date ML models cut inference time by -70 % and unlocked predictive market-trend scoring.
Faster releases: CI/CD and clear docs halved time-to-feature from fortnightly to weekly sprints.
Stronger retention: early adopters reported +18 % session duration and fewer manual data-cleanup tasks.
Future-proof stack: containerised workloads run on-prem for GDPR compliance yet are ready for hybrid cloud if needed. Together these wins cemented the platform’s reputation as the go-to AI property data tool in the UK estate-agent market
30+ production-grade data pipelines now refresh proprietary datasets hourly, enabling new prospecting features for agents.
Up-to-date ML models cut inference time by -70 % and unlocked predictive market-trend scoring.
Faster releases: CI/CD and clear docs halved time-to-feature from fortnightly to weekly sprints.
Stronger retention: early adopters reported +18 % session duration and fewer manual data-cleanup tasks.
Future-proof stack: containerised workloads run on-prem for GDPR compliance yet are ready for hybrid cloud if needed. Together these wins cemented the platform’s reputation as the go-to AI property data tool in the UK estate-agent market