Building a verified landlord-tenant compliance library with AI research agents
A single rental notice can be governed by four layers of law at once, state, city, housing authority, and federal voucher rules, and they don't always point the same direction. Getting from that tangle to 239 verified notice-language requirements meant building a reference that would rather refuse an answer than guess at one.
Category
Compliance research and verification, across many jurisdictions at once.
What I was trying to accomplish
One current, source-verified reference that answers a simple question for any property: what a compliant notice has to say, when, and to whom, when state law, city rules, and housing-authority requirements all layer on top of each other.
The data challenges
The state landlord-tenant statute, ordinances for more than twenty governing jurisdictions, rules from about ten housing authorities, federal voucher regulations, and a property directory that had to be mapped to the jurisdiction that actually governs each building, not just the city on the mailing address. The research came together in three separate passes that did not always agree with each other.
The result
239 notice-language requirements, each labeled confirmed verbatim, paraphrased, or not found, with only verbatim-confirmed language allowed on an actual notice. A most-restrictive-rule-wins model resolves conflicts, and the compiler that assembles the reference refuses to produce output if a required section is missing rather than silently leaving a gap.
Technical details
Tools: Claude, parallel research agents, an independent verification pass, Python
A property management company with buildings spread across many Washington cities faces a problem most multi-jurisdiction operators share: the rules for a rent-increase, renewal, or delinquency notice change depending on where the building sits. State law sets a floor, many cities add their own requirements, and housing authorities layer on more for voucher tenants.
The research was split across ten parallel agents that pulled primary-source text, then a separate verification pass re-read the highest-stakes language directly from official government sources and compared it character by character. The research was done in three passes (city law profiles, renewal requirements, and a newer notice-language library), and the layers overlapped and sometimes disagreed, so the final step was a consolidation: one structure per jurisdiction, playbooks for assembling a compliant notice, a machine-readable record layer for a future notice generator, and a two-page quick reference for property managers.
The challenges were instructive. An early statute file had been silently truncated while its header claimed full coverage, which is why the compiler now fails loudly. One housing authority published conflicting notice periods across its own materials, so the library records the conflict and standardizes on the safer figure. Older statute sections carried forward from the first pull are labeled as not yet re-verified, and several items are gated for counsel review before operational use. This is an operational compliance reference, not legal advice, and it was built that way on purpose.
Pull up a chair.
Let's talk through where things stand, and where they could go.