Perspective Labs Inc. · Property intelligence
Delve
Making property listings and strata documents easier to investigate, with structured reports, calculations, and evidence behind the findings.
In development · working name
- ExtractListings & documents
- EvaluateRules & calculations
- ExplainFindings & evidence
What should a buyer investigate before making an offer?
The useful information about a property is scattered across listing descriptions, building records, and strata documents. I’m building a way to bring those sources into one report, surface questions worth asking, and show where the available information is incomplete.
A pipeline with distinct responsibilities
Site-specific extractors turn listing pages into structured fields and evidence snippets. Deterministic code handles property-risk factors and carrying-cost calculations. AI adds a written analysis and follow-up questions. These stages share a pipeline between the web application and its API.
Connecting a listing to building context
The implementation brings in building events, comparable properties, neighborhood information, and signals extracted from strata PDFs. Document analysis looks for subjects such as special levies, deferred maintenance, reserve funds, and insurance concerns. Reports retain extraction coverage, confidence, and citations alongside the findings.
Making incomplete information visible
A missing field is different from a clean bill of health. The code tests extraction, scoring, keyword detection, and calculations, including minimal listing input. The current document parser reads a bounded amount of text, so document completeness and citation quality need further evaluation before the reports can support a buyer’s due diligence.
Current stage
A prototype with listing extraction, scoring, report storage, and AI-assisted document analysis. I’m developing and evaluating the reporting workflow. Delve is a working name.
What I’m working toward
Evaluate representative listings and full document sets against manually reviewed findings, then test a focused reporting workflow with prospective users.