Tenant Advocate
Gemini 4 NYC HackathonTenard: your rights as a tenant, made simple The problem NYC renters living with unfixed repairs, like mould, leaks, peeling paint, missing window guards or dead smoke detectors, rarely know whether the problem breaks the law. They also can't prove it in a way that counts, and don't know what to do next. Many are afraid to push their landlord, so the problem stays. What it does Start a case: the tenant opens a chat, taps a problem or describes it in any language, and adds a photo or video. Check it against the law: Tenard says whether it's a possible violation of the NYC Housing Maintenance Code, cites the section, and gives the violation class and the landlord's deadline. When the answer depends on a fact the photo can't show (is there a child under 6, is it only on tile), it asks. Explain rights: it explains the tenant's rights in plain language, including protection from retaliation. Draft the next step: first a letter to the landlord, then a 311 complaint in the city's own categories. The tenant sends everything themselves. Track the case: each case follows the escalation path (notice → 311 → HPD inspection → violation → Housing Court) and builds the evidence a judge will want. It closes when the tenant confirms the repair with a photo. How it works Gemini 4 Argon proposes which problems a photo shows. The Decisions API confirms each one with a probability, reads facts like mould size, and checks photo quality. Code, not a model, picks the class and deadline from a 92-row decision matrix built from the Housing Maintenance Code and HPD's violation records. Argon writes the explanation from only the cited sections, and every quote is checked word for word. A Managed Agent (Gemini 4 Hydrogen) pulls the building's HPD record and builds a case file. Proof it works Test set: 175 photos and 22 video clips from public, licensed sources. The labels were frozen before any app code existed. Photos: Tenard made a third fewer wrong reports than a single Gemini call (27 vs 41) and fewer than half the false alarms on look-alike photos (3 of 24 vs 7 of 24). Video: 17 of 22 clips answered exactly right, against 14 for the baseline, with zero false alarms. Where it still falls short: it misses some real problems (85 of 98, against the baseline's 91). It's weak on leaks and on mid-sized mould. It says "the inspector decides" when the law doesn't settle the class. What's next Heat and hot water, pests, NYCHA and Section 8 routes, review of the decision matrix by a housing lawyer, and testing with real NYC tenants' photos. Tenard explains the law. It isn't a lawyer, and every result is a possible violation, not an official finding.





