Kindred

Inspiration Most unclaimed-money tools exist for one person checking their own name, once, if they remember to. But money doesn't just sit unclaimed for the person searching — it sits unclaimed for aging parents who don't know to look, for siblings who moved states years ago, for the whole extended family nobody's actively watching out for. We wanted to build something that treats "checking for unclaimed money" the way you'd treat checking in on family — ongoing, not a one-time favor to yourself. What it does Kindred searches California's public unclaimed-property records for a name and shows what's owed, in seconds. What makes it different is the Money Map — instead of a flat list of results, matches are shown as glowing markers across a timeline of the cities tied to your search, tappable to reveal the full match detail (who owes you, how much, and a direct link to file the claim on the real, official state site — Kindred never files claims on a user's behalf, and searching is always free). Search handles messy real-world names with nickname-aware fuzzy matching (so "Bob" correctly matches "Robert"), and when a common name returns many possible matches, Kindred surfaces a disambiguation view so users can narrow by city instead of guessing. Kindred Pro extends this from a single search to ongoing monitoring: saved profiles for family members, and automatic re-scans so nobody has to remember to check again. How we built it Frontend: React Native + Expo, with react-native-reanimated for the Money Map's motion (pulsing match markers, animated detail cards) and expo-haptics for tactile feedback on key moments. Backend: Python + FastAPI, indexing California's real public unclaimed-property bulk data (published directly by the California State Controller's Office) into SQLite, with rapidfuzz powering the fuzzy and nickname-aware name matching. Monetization: RevenueCat, handling the subscription entitlement for Pro (family profiles + re-scan alerts) and the native paywall UI. We leaned heavily on AI-assisted development to move fast — describing features and interactions in detail and iterating against real, running code rather than writing everything by hand — and treated automated + manual verification as a first-class part of the process, not an afterthought. Challenges we ran into Getting real government data in was harder than it looked — the state's public files don't come with a clean, documented schema, so ingestion needed real column-mapping and validation against the actual downloaded file rather than assumptions. Building the Money Map itself took multiple real iterations — an open-ended "make it visual and delightful" instruction is genuinely hard to get right on the first pass, AI-assisted or not, and we learned to be increasingly specific and to verify against the running app rather than trusting a description of what was built. We also had a real (and useful) scare with API key hygiene: RevenueCat separates secret keys from public SDK keys, and it's easy to accidentally copy the wrong one — we caught it, rotated the exposed key immediately, and it reinforced how seriously we now treat credential handling. Accomplishments that we're proud of A signature, genuinely interactive feature (the Money Map) rather than a plain results list — built, tested, and confirmed working end to end. Real government data, not a mocked dataset — verified against known fake sample records to confirm we were actually querying live public data, not our own test fixtures. A disambiguation experience for common names that we hadn't originally scoped, but added once we saw how often "Maria Garcia"-style names returned dozens of legitimate matches. A monetization model that's honest about what it charges for — Kindred charges for ongoing monitoring convenience, not for locating or filing a claim, which is explicitly always free through official government channels. What we learned We learned a lot about the unclaimed-property space itself — how fragmented state-by-state data access is, and how meaningfully California stands out by publishing its full dataset openly. We learned that AI-assisted building is genuinely powerful but requires real verification discipline: an AI tool confidently saying a feature is "done" isn't the same as it actually being done, and building small automated checks (like a script that confirms real vs. sample data programmatically) turned out to be one of the most valuable things we built. We also came away with a much sharper sense of how consumer fintech-adjacent products need to earn trust explicitly — through visible disclosures, not just a working feature. What's next for Kindred Expanding beyond California to additional states as their public data becomes accessible, prioritizing states with the largest populations first. A give-back mechanic tied to Pro subscriptions, granting free monitoring access to families facing financial hardship. Push notifications (via OneSignal) for real-time re-scan alerts instead of manual re-checks. Exploring on-device tools (like scanning old mail for potential unclaimed-property leads) as an additional, privacy-preserving way to surface matches without requiring manual search.