Pungu x - stock.adobe.com

News

Lendi Group standardises on MongoDB for AI-ready data layer

Following a merger that left the Australian fintech with a fragmented data architecture, Lendi Group has consolidated its databases onto MongoDB Atlas to reduce microservices sprawl and power AI-driven broker tools

Stephen Withers
By
Published: 17 Mar 2026 12:30

Home loan-centric fintech Lendi Group has standardised on MongoDB as the foundation for an operational data layer designed with artificial intelligence (AI) projects in mind.

The group was formed following the merger of digital disruptor Lendi and long-established mortgage broker Aussie Home Loans. Consequently, the combined business struggled with a fragmented data architecture inherited from its predecessors.

With over 500 deployable components, the environment was expensive to maintain and unsuitable as a foundation for the company’s aspiration to become an AI-powered provider of property search, buyer advocacy, mortgage broking, conveyancing and ownership tools.

That AI journey started in 2021, Lendi’s chief technology officer, Devesh Maheshwari, told Computer Weekly, with a plan to go beyond mortgage broking to support the entire home ownership experience. The idea was that “you don’t have to go anywhere else”, he explained.

By 2025, Lendi realised it needed to do something different to retain its position as a disruptor. Discussions with the board led to what became known as Project Aurora – an initiative to use AI to augment brokers’ expertise. However, it was clear that technological change was needed to support this plan, specifically the creation of an operational data layer to underpin such projects.

“Our developers were spending time tuning and maintaining our databases, rather than pushing features,” said Maheshwari. “To build an architecture that could support our AI-native vision, we needed to reduce our microservices sprawl, reduce complexity and look at consolidating core operational data into a unified data layer.”

After the first week of the project, the team decided to replace its previous mix of databases, including PostgreSQL, DynamoDB, an earlier deployment of MongoDB and others, entirely with MongoDB Atlas. “There simply wasn’t another option that offered the flexibility of the document model and the power of MongoDB’s integrated, AI-ready data platform,” said Will Hargan, senior AI systems engineer at Lendi.

Capabilities

Specific capabilities highlighted by Lendi included scalability, achieved through MongoDB’s horizontal sharding and local indexing; support for a “document-first” unified schema strategy; robust security and compliance features; and built-in vector search to support AI applications without having to deploy and maintain a separate vector database.

Details of 14 million properties now reside in the MongoDB database. While not all of the firm’s six million customer records have been migrated yet, Maheshwari expects that task to be completed by the end of 2026.

One of the goals of Project Aurora is to support deterministic processes with AI that can interpret unstructured data and automate workflows, enabling brokers to work more quickly. For example, a broker might want to ask three mortgage providers how much they would lend a particular client for a specific property. Manually transcribing all the necessary data takes a proficient broker at least 45 minutes, said Maheshwari.

But Lendi’s AI-powered system can submit those three requests in parallel, reducing the whole exercise to just two or three minutes. Apart from saving the broker’s time and relieving them of a tedious task, the process can be completed while the customer waits, which has been shown to increase conversion rates.

In addition, Lendi’s brokers now use AI to analyse sale contracts and bring any issues to customers’ attention. The significance of this is that most real estate action occurs at weekends when conveyancers aren’t at work. This immediate guidance can improve a buyer’s bargaining power ahead of getting their conveyancer’s formal opinion, said Maheshwari. He added that while it does not eliminate the role of the conveyancer, it delivers a good experience for the customer.

Read more about AI in Australia

Governance is an important part of the development of both the operational data layer and the AI projects being built upon it. Traditional software engineering practices remain important, said Maheshwari, with risk and compliance specialists involved throughout the process.

For example, Lendi created its own toolchain to test different large language models against the same prompts, and to provide continuous monitoring of agent performance. “We have complete explainability and traceability of these models and their behaviour,” he said.

The architecture also allows the company’s developers to work more quickly. For instance, the Lendi Guardian home and loan companion that lets customers track rates and equity, and refinance with a single tap, was delivered in just 12 weeks. This is about 40% faster than would have been possible under the previous microservices-centric approach. From a business perspective, Guardian is showing “a substantial two-digit improvement” in the proportion of applications that reach the lodgement stage.

Looking ahead, Lendi plans to move beyond AI-powered automation to a situation where AI agents handle routine processes autonomously, freeing human brokers to concentrate on complex lending structures and the human relationship aspects of the business.

Read more on Database management