An inbox full of receipts. A clearer picture.
An email-to-insights product that turns receipts and invoices into structured financial transactions, with resilient sync and a companion mobile experience.
Your receipt.
Northside Studio
Software & subscriptions
−$24.00The challenge.
Receipt emails vary by merchant, format and language. The same purchase can appear in several messages, providers impose quotas, and a large inbox cannot depend on one uninterrupted request.
What I built.
I built Airtail across the web interface, email integrations, transaction pipeline and mobile API. Its architecture combines deterministic extraction for known patterns with model-based handling for ambiguous messages.
- ↗︎Rules-first extraction with model fallback
- ↗︎Resumable, chunked email sync
- ↗︎Web product and companion iOS app
The decisions behind it.
Use a model where it adds value
Rules handle recognisable receipt patterns before the AI extraction stage. The ambiguous remainder goes through a model fallback chain with bounded output and parsing.
Resume instead of restart
Chunked sync jobs use checkpoints, provider cursors, heartbeats and retries. A provider failure or quota limit should not force an entire inbox to start again.
Separate identity from access
Application authentication and email-provider access are distinct concerns. Provider-specific integrations, user-associated records and deduplication support the data flow.
What matters most.
The best AI pipeline is not necessarily the one that calls a model most often. Deterministic work, resumability and careful validation make the product more dependable.
Airtail is an independent project with private source code. Interface illustrations and the portfolio lab use synthetic data. The lab is a standalone demonstration of the pattern, not Airtail’s production pipeline.