A Singapore company, born in a university research lab, builds domain-specific AI that turns a bank's messiest paperwork - trade invoices, financial statements, letters of credit - into decisions.
A corporate loan officer at a mid-sized Asian bank spends a surprising amount of time doing something no one went to business school for: reading. Financial statements in one format, trade invoices in another, a letter of credit written to rules older than the internet. The judgment call at the end takes minutes. The reading takes hours. 6Estates built a company around that gap.
Founded in Singapore in 2014, 6Estates makes software that reads the documents banks would rather not. It extracts the numbers, checks them against the rules, and hands a human the structured result. The pitch is narrow on purpose - not general artificial intelligence, but domain-specific AI aimed squarely at finance, in the many languages of Southeast Asia.
Most startup myths involve a dorm room. 6Estates has a footnote instead: it spun out of NExT++, a joint research center run by the National University of Singapore, Tsinghua University in China, and the University of Southampton in the UK. The founders, Huanbo Luan and Roger Yuen, came out of that world - natural language processing, machine reading comprehension, the academic machinery for teaching computers to understand text.
The move that mattered was leaving. A paper about machine reading is worth very little to a bank until it can clear a real letter of credit. Dr. Luan, who serves as Founder and CEO, pointed the research at the least glamorous corner of finance: the paperwork.
6Estates sells to financial institutions - more than 30 of them across Asia. The named list reads like a regional banking directory: Bank Danamon, Adira Finance, PT Bank HSBC Indonesia, Bank Central Asia, plus newer lenders like Broom and Nikel. These are institutions that process thousands of documents a day and cannot afford to get a number wrong.
The problem 6Estates solves is deceptively boring. The bottleneck in lending is rarely the decision - it is the data entry that precedes it. A credit analyst who spends three hours keying a financial statement into a model has three fewer hours for the judgment only a human should make. Automate the reading, and the decisions flow.
The scope has widened over time. The early customers were banks, but the same document problem shows up in shipping, logistics, and insurance - anywhere a business runs on forms that arrive as scans, photographs, and PDFs rather than clean database rows. When 6Estates raised its 2022 round, those adjacent sectors were named explicitly as the next targets. The bet is that a system good enough to read a corporate balance sheet is also good enough to read a bill of lading or an insurance claim, once it learns the vocabulary.
The core engine is IDP, Intelligent Document Processing, which unlocks complex unstructured documents. Around it sit the finance-specific tools: FSA, a Financial Statement Analyzer that automates corporate credit reading; FAAS, a Finance-as-a-Service layer that runs an end-to-end lending workflow; and LC Automize, which handles trade finance.
LC Automize is the clearest example of the approach. A letter of credit is checked against two rulebooks the trade world has used for decades - the UCP (Uniform Customs and Practice for Documentary Credits) and the ISBP (International Standard Banking Practice). 6Estates encoded those rules into software, combined AI with robotic process automation, and let the document effectively check itself. Bank Central Asia partnered with the company to build exactly this kind of AI-plus-RPA trade finance solution.
More recently the company pushed both up and down that stack. Up: an Agentic AI Platform of modular agents for end-to-end loan processing and risk assessment. Down: a GPU Cloud and an LLM Token Factory - infrastructure aimed at Southeast Asian enterprises that want to run models without shipping their data overseas.
The document-AI field is crowded - Docsumo, fileAI, Hyperscience, Rossum, and a bank's own in-house automation team all compete for the same work. 6Estates draws its lines in three places.
First, domain. It builds finance-specific, multilingual, multimodal models rather than a general engine, on the bet that when a wrong number means a defaulted loan, precision beats generality. Second, language. Southeast Asia's hundreds of languages and thousands of document formats are treated as a moat, not a nuisance. Third, privacy. 6Estates deploys on-premise for banks that legally cannot send documents to the cloud - a feature that closed deals cloud-only vendors could not.
For a company of roughly 41 people, the partner roster is unusually heavy. 6Estates lists NVIDIA for AI infrastructure, Microsoft Azure for cloud, AI Singapore as a national programme partner, and consulting and integration names including KPMG Japan, IBM Consulting, and banking-software specialist Silverlake. In Indonesia, ICT provider Lintasarta helps carry deployments into the region. The pattern is consistent: 6Estates supplies the finance-specific intelligence and leans on partners for reach and horsepower.
The research bloodline still shows in how the company talks about itself. It runs an LLM Open Course and an AI Advanced Workshop, and frames its offering as a full stack - hardware, base models, tuning, and applications. That is an unusual thing for a document-processing vendor to say, and it hints at the ambition underneath the paperwork: not just to read documents, but to own the machinery that makes reading them possible in Southeast Asia.
One product wanders furthest from the core. FundFluent, launched in Hong Kong, is an SME funding platform - a move from selling tools to banks toward sitting closer to the borrower. It is a small experiment against a large gap: small and medium businesses across Asia remain chronically underfunded, and a company that already understands their financial documents is well placed to help match them with capital.
The model is straightforward B2B enterprise software. Revenue comes from recurring SaaS subscriptions and on-premise licenses, layered with GPU cloud and LLM infrastructure services, integration contracts, and training programs like the LLM Open Course and AI Advanced Workshop. Estimated annual revenue sits around USD 6.5 million.
The funding history tracks the regional strategy. An early Series B drew GDP Venture (part of Indonesia's Djarum Group) and Central Capital Ventura (the venture arm of Bank Central Asia). In February 2022 the company closed a USD 6.2 million Series B+ led by Sinarmas Group, with Enterprise Singapore's SEEDS Capital and Farquhar VC - money earmarked to push IDP deeper into banking, trade, shipping, logistics, and insurance.
6Estates sits at an unfashionable but durable spot in the market: the layer between a document and a decision. It is not trying to be a consumer app or a foundation-model lab. It is building the plumbing that lets a bank in Jakarta or Singapore run domain AI on its own terms, in its own languages, inside its own walls. That is a smaller ambition than "artificial general intelligence." It is also one a loan officer, three hours of reading lighter, would happily pay for.