Sao Mai Labs

Running daily at

An Phat Accounting
Hai Au Contact Centre
Thinh An Insurance
Minh Long Audit
Cang Xanh Logistics
Moc Lan Retail
Van Xuan Legal
Hong Linh Clinics

3.1Mpages read every month

99.4%field-level accuracy on invoices

0.8saverage time per page

4,200hof speech from all 34 provinces

Products

Two models, one job: understand real Vietnamese.

Real Vietnamese has rushed diacritics, stamps printed over handwriting, and callers who say "răng" instead of "sao". General multilingual models stumble on exactly those cases. We train specifically for them.

Mai Read

Documents

Photographed on a phone or scanned on a copier, every page comes back as structured fields with a confidence score, so your team knows exactly which ones need a second look.

Recognises 37 common document types: VAT invoices, citizen ID cards, land-use certificates, contracts, handwritten receipts
Cross-checks the amount in figures against the amount written in words
Returns JSON through a REST API or posts straight into your accounting software

One day at An Phat Accounting, Ha Dong14 Sep 2026

412invoices read automatically
This used to take two accountants the whole morning.
9invoices flagged for review
Confidence below 95%, mostly blurry photos or stamps over figures.
2amount mismatches caught
Words and figures disagreed. Manual entry had missed both.

AutomaticFlaggedMismatch

Mai Listen

Voice

Live call transcription that normalises regional dialects into standard Vietnamese, then tags intent and sentiment so the contact centre knows which calls to handle first.

Around 300 ms latency, fast enough to suggest replies while the caller is still talking
Trained on 4,200 hours of recordings covering Northern, Central and Southern accents, including Nghe Tinh, Hue and Quang Nam
Runs on your own servers, so recordings never leave the building

Call #88213 · Customer care00:18

00:03

"Răng mà hôm qua tui gọi mãi không ai nghe hết rứa?"

Normalised: Sao hôm qua tôi gọi mãi không ai nghe vậy?

"Why did nobody pick up when I kept calling yesterday?"

00:09

"Cái thẻ tê hắn trừ tiền hai lần, chộ không?"

Normalised: Cái thẻ kia nó trừ tiền hai lần, thấy không?

"That card charged me twice, see?"

Accent: Nghe An Intent: double-charge complaint Sentiment: frustrated Priority: high

Word error rate (WER) by accent region

Lower is better. Measured on a 12-hour test set of real contact-centre calls, August 2026.

Mai ListenPopular multilingual model

0%10%20%30%40%

Solutions by industry

Built for the paperwork and phone lines Vietnam actually runs on.

Banking

eKYC from citizen ID cards, loan files and handwritten branch forms, read in seconds.

-72% loan file intake time

Accounting

VAT invoices, receipts and bank statements posted straight into the ledger, with mismatches flagged.

140 firms in production

Contact centres

Live transcripts for every accent, auto-tagged intent and sentiment, and an angry-caller queue.

-46% average handling time

Insurance

Claims packs with handwritten hospital bills and prescriptions, sorted and checked before an assessor opens them.

3.4 days faster claim decisions

Security & data residency

Your documents stay in Vietnam. Your recordings can stay in your building.

Hosted in Vietnam

Production servers in Hanoi and Ho Chi Minh City data centres. No offshore processing.

On-premise option

Both models ship as containers that run on your own hardware, fully offline.

Encrypted end to end

TLS 1.3 in transit, AES-256 at rest, and keys you can rotate yourself.

Never trained on your data

Customer data is excluded from training unless you agree to it in writing.

Journey & funding

Almost two years on revenue alone, then venture capital.

We were selling before we had investors. Below is every dollar we have raised, including the round where the committed amount and the amount received were not the same.

22 months bootstrapped · Mar 2022 – Dec 2023$122.7M received · since Jan 2024

Mar 2022

Founded · Bootstrapped

VND 600M

Personal savings of the three co-founders

Three people in a rented room in Cau Giay. The first paying customer arrived eight months later: an accounting firm in Ha Dong, at VND 2 million a month.
Sep 2023

Break-even

$0 outside capital

Revenue covered the salaries of a seven-person team

We only took investment once Mai Read had 60 customers paying every month.
Jan 2024

Seed

$1.2M

Northpeak Ventures + Atlas Foundry + Horizon Labs

Used to build Mai Listen and open the Ho Chi Minh City office.
Dec 2024

Series A

$8.5M

Vertex Capital + Nova Ventures + Arcadia Partners
May 2025

Series B

$95M

Summit Ventures + Northstar Capital + Meridian Growth
Nov 2025

Next round · Committed

$120M committed

Strategic ecosystem investors + existing backers
Feb 2026

UPDATE
Next round · Received

$18M ~~$120M~~

The amount received was far below the commitment. We publish the actual figure so investors and employees have the same, accurate picture.

Customers

What changes when the paperwork reads itself.

Two accountants used to spend the whole morning keying in invoices. Now Mai Read finishes 400 of them before 9 a.m., and we only check the handful it flags.

TTPham Thu TrangChief Accountant, An Phat Accounting

New agents struggled with callers from the central provinces. Mai Listen transcribes and tags the complaint while the call is still live. Handling time dropped by almost half.

VHLe Van HungHead of Operations, Hai Au Contact Centre

Our requirement was simple: claims data does not leave our building. Sao Mai installed both models on our own servers in two weeks.

MANguyen Thi Mai AnhHead of IT, Thinh An Insurance

4h → 25mdaily invoice entry at An Phat Accounting

How we work

Three rules we have kept since day one.

Sell first, build second

Every major feature starts with a customer who has agreed to pay for it. Mai Listen exists because a contact centre asked for it.

Measure errors, not demos

Every demo looks good. We publish error rates on real data, including the accents where we are still weak.

Data stays in Vietnam

Servers are in the country. Customer data is never used for training without written consent.

Leadership

120 people, two offices, one language.

Co-founder · Chief Executive Officer

Nguyen Minh Khoa

Based in: Hanoi headquarters
Education: B.Eng. Computer Science · MBA, Technology Management
Experience: 12 years in product, 8 of them in fintech
Legal representative: Sao Mai Labs JSC, since March 2022
Board: Chair, Sao Mai Labs JSC

"We knocked on the door of one accounting firm in Ha Dong and asked what wasted their mornings. Everything we build still starts with that question."

Khoa leads strategy, fundraising and enterprise sales. He closed Sao Mai's first customer himself, led the Seed through Series B rounds, and still visits customers every week. Before Sao Mai he ran product for digital lending and payment platforms serving more than two million Vietnamese users.

2022 – now
Co-founder & CEO, Sao Mai LabsGrew the company from three founders to 120 people and two offices.
2018 – 2022
Head of Product, digital lending platformLaunched instant loan approval; cut document review from three days to four hours.
2014 – 2018
Product Lead, mobile payments startupShipped QR payments to 40,000 small merchants.

Enterprise salesFundraisingData privacy policyProduct strategy

Tran Thu Ha

Co-founder · CTO

MSc in natural language processing. Built Sao Mai's first Vietnamese handwriting dataset from 40,000 paper receipts and leads all model research.

Le Quang Vinh

Co-founder · Head of Speech

Grew up in Nghe An, so he knows why machines mishear "răng", "rứa" and "chộ". Owns Mai Listen end to end.

Where the other 117 people work

Headcount, October 2026

Machine learning

Sales & deployment

Vietnamese linguists

Infrastructure

Operations

Careers

We are hiring for three roles.

Send your CV and one piece of work you are proud of to the email below. We reply to every application within five working days.

Machine Learning Engineer, Speech
Hanoi · Full-time
Vietnamese Corpus Linguist
Hanoi · Part-time
Enterprise Solutions Engineer
Ho Chi Minh City · Full-time

Contact

Send us the 20 hardest-to-read pages you have.

Within 48 hours you get Mai Read's output back with the error rate measured on those exact pages. Then try 1,000 pages free.

lienhe@saoai.id.vn

Sales, partnerships, press and job applications. We answer within one working day.

HanoiHQ

Open now

Floor 12, Sao Mai Building
86 Duy Tan Street, Cau Giay Ward
Hanoi 100000, Vietnam

Mon–Fri 08:30–17:30 · Sat 08:30–12:00Local time --:--:--

Ho Chi Minh City