PRODUCT ARCHITECTURE

How Trayd turns a tradesperson's chat into an invoice — automatically.

Three layers: a chat-first capture surface, an AI assembly engine, and an employer dashboard. The AI does the work; humans only confirm.

Core flow

GPS arrival

Chat capture

Photos

Receipt OCR

AI assembly

Employer review

Invoice PDF

Paid

System layers

Capture (PWA, offline-first)

· Chat UI · taps + voice
· GPS auto-detect customer
· Camera + receipt OCR
· Local IndexedDB queue

AI Engine

· Gemini 2.5 Flash — extraction
· Claude Sonnet — reasoning
· Vision OCR for receipts
· Confidence scoring + fallbacks

Backend

· Postgres + Row-level security
· Object storage for photos
· PDF generation (server)
· Webhooks for sync

Pricing & rules engine

INVOICE FORMULA

labour     = Σ (employee.hours × employee.rate)
materials  = Σ receipts.items + manualMaterials
subtotal   = labour + materials
vat        = subtotal × 0.23     // Ireland
total      = subtotal + vat

CONFIDENCE GATES

✓ > 0.85 — auto-fill, employer review
! 0.6–0.85 — flag line item, ask employee
✕ < 0.6 — block until confirmed

Data model

tenant ──┬─< user (employee | employer | admin)
         ├─< customer ──< job ──┬─< receipt ──< receipt_item
         │                       ├─< material
         │                       └─── invoice ──< invoice_line
         └─< rate_card