Environmental · LLM

Water Advisor AI

Address → local US EPA Consumer Confidence Report → LLM-parsed water quality advice with per-parse cost tracking.

Tech Stack:Next.jsTypeScriptPostgreSQLOpenAI

The Problem

US municipalities publish Consumer Confidence Reports (CCRs) annually under EPA mandate, but the reports are long, inconsistently formatted PDFs that ordinary consumers cannot act on. A platform that turns 'address' into 'concrete water-quality advice' solves a real public-health information gap, as long as the parsing layer can be trusted and audited.

The Solution

Apex36 built an address-to-CCR lookup engine that resolves a user address (or ZIP) to the responsible water utility's latest report, parses the CCR PDF via LLM into structured PostgreSQL records (per-contaminant rows with units and limits), and returns a consumer-readable summary with recommended treatment guidance. Every parse records model used, input tokens, output tokens, and USD cost with six-decimal precision for auditability.

Features

Address → CCR lookup

Resolves a user address or ZIP to the responsible water utility's latest EPA Consumer Confidence Report.

LLM parsing of CCR PDFs

Parses unstructured PDFs into structured PostgreSQL records, per-contaminant rows with units and limits.

Consumer-readable summary

Plain-language summary plus recommended treatment options, grounded in the parsed report.

Per-parse cost observability

Every LLM parse records model used, input tokens, output tokens, and USD cost with six-decimal precision.

EWG cross-reference

Cross-references the EWG Tap Water Database for additional verification.

Results / Impact

Phase 1 MVP shipped

in March 2026; Phase 2 in progress as of April 2026.

Per-parse cost instrumentation

model, input/output tokens, USD to six decimals on every call.

Model-agnostic parser

with full auditability, quality and economics can be audited per call.

FAQ

Converts a US address or ZIP into actionable water-quality advice by looking up the local utility's EPA Consumer Confidence Report and parsing it with an LLM into structured, human-readable guidance.

Every LLM parse records model used, input tokens, output tokens, and USD cost with six-decimal precision, so quality and economics can be audited per call.

Ready to build something impactful?

Let's discuss your project and how we can help you ship faster and smarter.

Book a Free Strategy Call