AutoQuote · CMS-AQI
Multi-Tenant Document AI · 12 Specialised Agents
Multi-tenant document AI platform. Ingests invoices, receipts, PODs, credit notes, quotations, subscriptions, and bank statements via email or upload, classifies them, extracts structured data, and stores results.
The Brief
Problem
Construction clients were drowning in PDFs — supplier invoices, PODs, credit notes — all manually keyed into spreadsheets. Rebate eligibility and reconciliation took weeks per month.
The Architecture
Decision
12 specialised Gemini agents, each handling one document type. Three-tier classification (keyword → Gemini → Agent 1 deep). Per-client sequential queue, cross-client parallel up to 20 simultaneously. Wasabi storage, MySQL data, SSE for real-time progress in the dashboard.
The Outcome
Result
Documents flow from email to dashboard in seconds. Rebate calculations auto-run. POD linking and bank reconciliation reduced from days to minutes.
How it actually works in production.
Ingest
Email (IMAP)
multi-mailbox
Manual upload
dashboard
Wasabi S3
fast upload
Processing queue
MySQL · per-client
Classify (3-tier)
Keyword match
fast path
Gemini classify
mid path
Agent 1 deep
fallback
12 Specialised Agents (Gemini)
Invoice
A02
POD
A03
Quotation
A04
Credit note
A05
Receipt
A06 · 95+ fields
POD linking
A07
Rebate × 3
A08a/b/c
Bank recon
A09
Generic
A10
Reclassify
A11
Subscription
A12
Persist & Notify
MySQL
auto_invoices, auto_pods…
Wasabi final
{client}/{type}/{file}
SSE event
live dashboard
Job complete
audit log
Animated · Built in code · No GIFs
See AutoQuote · CMS-AQI in action.
Demo video
Email → 12-agent pipeline → MySQL (full demo)
~ 1:20 · coming soon
Image slot
Dashboard — live processing stats
coming soon
Image slot
Classifier with extracted invoice fields
coming soon
Image slot
Rebate manager — auto-eligibility check
coming soon
Image slot
Bank reconciliation — auto-matched lines
coming soon
Stack