// screenshots
data pipeline · open source
Full historical ingest of HM Land Registry Price Paid Data, enriched with EPC energy certificates for floor area and £/m². Postcode geocoding, Grafana dashboards, and a conversational analytics agent a local LLM drives over MCP — kept current with automated monthly updates.
A full historical ingest of HM Land Registry Price Paid Data — every recorded residential sale in England and Wales since 1995, ~28 million rows — loaded into Postgres and geocoded to coordinates from OS Code-Point Open for regional and map-based analysis.
On top sits an EPC (Energy Performance Certificate) layer: ~23 million domestic certificates address-matched to sales — Land Registry has no UPRN, so the join is built on normalised address keys — adding floor area, £ per m², and energy ratings to ~68% of transactions (88–92% in recent years). Grafana dashboards drive exploration, and a local LLM powers both a monthly market summary and a conversational analytics agent shaped around what a first-time buyer actually asks — what fits my budget, is this good value, £ per m², where's the market moving — served over an MCP server with every figure coverage-gated and grounded in real numbers.