AI & Automation · Document AI
Invoices, receipts and reports, read into your accounts.
Crumpled, tilted, three languages. Every field cross-checked by a second model, a person confirms the crop, and nothing has to leave your building.
- 4 of 5 receipts perfect
- ~95% field accuracy
- −80% manual entry
- EN/FR/ES
One receipt, start to finish.
- 01Photo
- 02Straighten
- 03Confirm crop
- 04AI reads
- 05Clean data
From case 20, Receipt Reader. A person nudges or accepts the crop before anything is read.
What it reads.
Invoices and receipts
Into Sage, Dynamics, QuickBooks or your ERP.
Company reports
Into summaries and Urdu audio briefings (Briefcast).
Exam papers
Into marks against the official scheme.
Product data
Classified and de-duplicated at scale.
On your own machines.
Document models run on a local GPU with open models, so there's no cloud API at read time.
$15K a year saved
A North American finance app reads English, French and Spanish scans on local GPUs, checks every result with a second model and posts to Sage and Microsoft Dynamics. That cut its cloud AI spend by $15,000 a year.
Case 12.
When a document doesn't read cleanly.
- A low-confidence field
- Flagged for a person.
- A missing total
- Held, not posted.
- An unreadable scan
- Returned with a reason.
Document AI we've built.
- Finance software
Invoice and receipt reading on local GPUs
$15K a year saved on cloud AI ·
- Document AI
Receipt Reader: nothing leaves the machine
4 of 5 real receipts read perfectly; the fifth off by 1 rupee · receipts and invoices · 2025 to 2026
- Document AI
An invoice reader that cuts manual entry by 80%
~95% field accuracy · Zynthrix · 2025 to 2026
- Document AI
Briefcast: company reports as Urdu audio briefings
~186 documents processed on the production server · 2026
- EdTech AI
Marking handwritten exam papers with AI
451 marking-scheme PDFs registered · prototypes by Anees and the Zynthrix team · Mar to Apr 2026
- Market data
Cleaning product data for millions of e-commerce items
Millions e-commerce products · NIQ (Etilize) · 2009 to 2024
How it's priced.
- Sprint $490: we run your last 50 invoices through a reader and show the results.
- Build from $2,500.
- On-premise setup scoped separately.
- Tuning from $500/mo.
Prices exclude taxes and third-party costs (hosting, carrier minutes, AI usage), billed at cost to your own accounts.
See all pricesQuestions about document AI.
Can it run on our servers?
Yes. Open models can run on your own GPUs, so nothing leaves the building. We've done this for invoice reading, daily compliance checks and a receipt reader.
What does a call or a document cost?
Carrier minutes and AI tokens are billed by the providers to your own accounts at their published rates. We add no platform fee. Typical: ~$0.03 per call-minute on Ava; cents per document for reading; near zero for on-premise.
Which languages?
Our voice agents speak English and Urdu today, and our invoice reader handles English, French and Spanish. Tell us yours and we'll test it in a Proof Sprint before you commit.
How do you test it before it goes live?
On labelled test sets built from your own data, with automated evaluations, then staged UAT with the people who will use it. High-stakes actions stay behind approval gates after launch.
Send us one job.
Tell us what you need built. You'll have scope, price and dates in writing today, the work in 3–5 business days, and you pay only if you approve it.
Prices exclude taxes and third-party costs (hosting, carrier minutes, AI usage), billed at cost to your own accounts.