Why not use subscription-based replenishment?

The gap between purchase date and depletion date varies by household and product, so fixed shipment cadences are a poor fit. Subscription models solve the problem bluntly by shipping on a schedule regardless of actual consumption. Axis models real consumption from receipt history instead, so reorder timing matches how the household actually uses the product.

What is the current status of Axis?

Plans 1 (foundation), 2 (catalog), and 3 (ingestion and parser) shipped to production as of 2026-04-18. Real forwarded receipts have been validated end to end with two matched Observations confirmed on staging. Plan 4, which adds prediction and a one-tap reorder approval card, is drafted. Production URL is axis.aethelforge.ai.

Axis — iSimplifyMe Labs

Q: How does Axis ingest purchase data?

Gmail polling runs every thirty minutes via a Cron function, fetching new receipts and passing them to a Bedrock Haiku parser. The parser extracts product names, quantities, and purchase dates from email HTML. A match verifier checks each extracted item against the catalog using fuzzy matching, and confirmed matches write as Observation records in DynamoDB.

Abstract

Axis is a predictive replenishment product that passively monitors purchase history from forwarded email receipts, matches purchases to a curated catalog of household consumables, and surfaces replenishment suggestions before the product runs out. No manual logging — the data comes from receipts already in the user's inbox.

Problem

Household consumables (supplements, personal care products) run out on irregular schedules that are hard to track manually. Subscription models solve the problem bluntly by shipping on a fixed cadence regardless of actual consumption rate.

The gap between purchase date and depletion date varies by household and product, making fixed schedules a poor fit.

The signal that tracks real consumption already exists: purchase receipts in Gmail. The problem is extracting structured product and quantity data from unstructured email HTML, matching it to a known product catalog, and using that match history to build a consumption model per user per product.

Approach

Ingestion pipeline

Gmail polling (every 30 minutes via a Cron function) fetches new receipts and passes them to a Bedrock Haiku parser. The parser extracts product names, quantities, and purchase dates from email HTML, then a match verifier checks each extracted item against the catalog using fuzzy matching.

Confirmed matches are written as Observation records in DynamoDB.

Catalog and foundation

The catalog is defined as JSON schemas in the repository and synced to S3 and DynamoDB via a GitHub Actions workflow (`catalog-sync`). Catalog validation runs as a separate CI check on every push.

The foundation uses Auth.js v5 with passkey authentication and a per-tenant DynamoDB structure, deployed on AWS via SST v3.

Prediction

Plan 4 (not yet shipped) will add prediction: given N observations on a product, estimate the next depletion date and surface an approval card for one-tap reorder.

Status

Plans 1 (foundation), 2 (catalog), and 3 (ingestion + parser) shipped to production as of 2026-04-18.
Two brands in the catalog (Thorne and Hiya); Gmail polling and Bedrock Haiku parser are live.
Real forwarded receipts validated end-to-end: 2 matched Observations confirmed on staging.
Production URL: axis.aethelforge.ai.
Plan 4 (prediction + approval + execution) is drafted; prerequisites: Cloudflare Workers Paid plan for durable objects.

Axis

What is Axis?