What changed here?
Build one answer across assessments, recorded sales, permits, violations, licenses, abatements, and demolition activity.
Active system
Philadelphia property intelligence
AI that can work across the public record.
Philadelphia already publishes deep property data. The value is adding an AI research layer that can move across those sources, compare records, and return a bounded answer with the city evidence attached. The joined reports make that possible. They are not the endpoint.
What the AI can answer
Property and block scaleHow does it compare?
Compare an address with its block across ownership, tax treatment, building mix, value history, and recent market activity.
Where did the answer come from?
Keep the city source, record date, and the boundary between a recorded fact and an interpretation attached to the answer.
Records become tools for an AI researcher
Address to cited answerResolve the place
An address resolves to the current parcel and hundred-block, with nearest-property handling for map taps that land beside a parcel.
Assemble the record
Assessment history, deeds, permits, violations, taxes, licenses, parcel geometry, and Census context become one bounded property and block record.
Let the agent query
A tool-calling AI agent selects the relevant Philadelphia sources for the question instead of treating the rendered report as a wall of text.
Return the evidence
The interface shows the research steps and returns a compact answer with citations to the records used.
The report is the AI's working context
Portfolio-safe reconstruction
The report organizes the records. The AI layer makes them useful as a research system. One question can cross deeds, permits, violations, licenses, taxes, assessments, nearby activity, and Philadelphia code without asking the reader to know which city system holds each part.
Cached property and block insights use a narrower path. The model receives a compact fact sheet, and every number in its answer is checked against that input. If the grounding check fails, the insight is not shown.
The project remains anonymous here. There is no live link, product name, private route, or account surface on this site.
The AI has to show its work
Inference boundariesA recorded owner name does not establish beneficial ownership. An absentee classification does not prove where a person lives. Missing records stay missing instead of becoming a zero. Research answers cite the source used for the claim.
The system keeps those distinctions visible in the report and in the generated answer. The agent can prioritize and explain the record, but it still has to show the line between a city fact, a derived comparison, and an interpretation.