The user manual your house never came with. Hearth turns photos, documents, and public records into the maintenance plan, inventory record, and historical archive your home has never had.
Hearth publishes a water quality profile for each city it serves, and these two are open to anyone, no account needed. Every number on them comes from EPA's Safe Drinking Water records and the utility's own published annual water quality report. Hearth reads those records the same way it does for the households it watches over, and shows its work. The tone is awareness, not alarm: it is the same report the utility already mails out, made easier to understand.
A groundwater system serving about 190,000 people. Five years of EPA violation history, the most recent lead and copper sampling, forever chemicals (PFAS) monitoring, and every contaminant the utility's annual reports list as detected, with the trend across the reports on file.
hearth.toddtech.llc/water/kalamazoo-miThe neighboring city, on its own groundwater system. The same picture: EPA compliance record, lead and copper sampling, forever chemicals (PFAS) monitoring, and each detected contaminant shown against the limit its report measures it by.
hearth.toddtech.llc/water/portage-miSources on every page: EPA's Safe Drinking Water Information System for violations and lead and copper sampling, and each utility's published annual water quality report, the Consumer Confidence Report it sends to customers, for everything detected in the supply. Hearth adds the plain-English reading and nothing else.
The home view: an AI-generated architectural sketch of the property, the facts pulled from public records, an emergency video reference panel, and a live preview of the habitat findings. Every tile is real data, updating as background workflows finish.
The first run narrates its own work as it happens: public records first, then radon, Superfund proximity, flood zones, and water quality. Showing the work builds more trust than a spinner ever could.
A photo of an appliance nameplate becomes a structured inventory record, with a review step between the extraction and the save. The same uploader takes multi-page service receipts.
Each appliance gets its own page: the extracted facts, an on-demand “Research this model” panel that streams a grounded summary of service life and maintenance, and a manufactured-date tile decoded from the serial number.
The finding for a nearby Superfund site: distance tier, NPL status, and EPA's Community Involvement Coordinator for the site. Every contaminant listed carries a plain-language description of what it actually is.
The compliance picture for the home's own water system: violation history, the contaminants detected against their federal action levels, and the actions Hearth recommends for this specific source.
The water module's real output, built from the home's own 2024 Consumer Confidence Report: every contaminant actually detected in the supply, mapped against six filter technologies, detected rows highlighted. Hearth reads that grid into one specific recommendation, with certifications named and install and annual costs attached, so it points at a product the homeowner can buy rather than a generic “consider a filter.”
A real maintenance plan for one appliance, tiered by urgency and scheduled by cadence: interval, seasonal, one-time, per-use. Every task carries its own reasoning.
Every task is one tap from a “why this task” card. Here a dishwasher-cleaning cadence tightened from the usual one to two months down to monthly, because the water-quality module found hard water in the county, and the card says exactly that.
JSONB columns earn promotion to real columns only when a query pattern demands it. The inventory and document tables both carry a `metadata` JSONB column for subtype-specific fields that don't need cross-row queries yet, things like vehicle VINs and plate states, pet microchip numbers, or a receipt's vendor and line items. When a query pattern finally needs one of those fields as a real column, say, aggregating values for an insurance valuation, that shape migrates out of JSONB into a column of its own. Until then, JSONB keeps the schema small and the path to a new feature short.
Workflow errors are sorted into terminal vs. retryable. Hearth's durable jobs run on Vercel Workflows, where a naive policy retries every failure as if it were transient. Hearth instead marks its orchestrator-internal errors as non-retryable: constraint violations, missing entities, a failed write. Those surface a deterministic bug in about two seconds rather than grinding through roughly thirty seconds of pointless retries, while genuinely transient errors like a flaky EPA call or a gateway hiccup still retry. That one distinction turns the workflow layer into a fast feedback loop during development instead of a latency tax.
Reads appliance nameplates, multi-page service receipts, and identifying documents. Haiku for fast, cheap structured extraction; every output is bound by a Zod schema, with unit tests pinning the no-prompt-leak contract.
A web-grounded summary of an appliance's service life, maintenance needs, and known issues, routed through the Vercel AI Gateway. It streams section by section, so there is real content on screen within seconds.
Decodes a manufacture date from the serial number when no install date is on file. The model has to name its rule, apply it, and return null rather than guess; only high-confidence decodes persist.
A durable background job turns the appliance's research summary, the home's habitat findings, and its receipts into a scheduled plan. Each task is stored with its own cadence, anchor, modifiers, and reasoning.
Mapbox address autofill feeds coordinates into EPA Envirofacts, FEMA NFHL, EPA's water-utility layer, and SDWIS. Shared per-state caches spread the cost of slow federal APIs across everyone in the same area.
Every habitat finding emits a step-by-step log of what was queried, what came back, and how severity was decided. The classification rules themselves are published on an in-app methodology page.
I'm available for contract and full-time roles, and happy to walk through any of these systems in detail.
Get in touch