How Lendi revamped the refinance journey for its prospects utilizing agentic AI in 16 weeks utilizing Amazon Bedrock

This publish was co-written with Davesh Maheshwari from Lendi Group and Samuel Casey from Mantel Group.

Most Australians don’t know whether or not their house mortgage remains to be aggressive. Charges shift, property values transfer, private circumstances change—but for the typical home-owner, staying knowledgeable of those adjustments is tough. It’s usually their largest monetary dedication, nevertheless it’s additionally the one they’re least outfitted to observe. And after they do resolve to refinance, the method itself calls for vital guide effort.

Lendi Group, one in every of Australia’s quickest rising FinTech corporations, acknowledged this hole and got down to remodel the house mortgage expertise by revolutionary expertise. By utilizing the generative AI capabilities of Amazon Bedrock, Lendi Group has developed Guardian, an agentic AI-powered utility that serves as an around-the-clock companion for householders, monitoring their loans, offering customized insights, and simplifying the mortgage refinance course of.

This publish particulars how Lendi Group constructed their AI-powered Residence Mortgage Guardian utilizing Amazon Bedrock, the challenges they confronted, the structure they carried out, and the numerous enterprise outcomes they’ve achieved. Their journey provides helpful insights for organizations that need to use generative AI to rework buyer experiences whereas sustaining the human contact that builds belief and loyalty.

Challenges

Lendi Group recognized a number of persistent challenges within the house mortgage journey that affected each prospects and brokers:

• Prospects struggled with restricted visibility into their mortgage place. Most householders lacked real-time insights into whether or not their present fee remained aggressive, how their fairness place modified as property values fluctuated, or how their general monetary well being impacted their mortgage choices. This info hole usually led to prospects lacking alternatives to save cash or make the most of their house fairness successfully.
• The refinancing course of was cumbersome and time-consuming. Even when prospects recognized higher charges, the paperwork and administrative burden of refinancing deterred many from appearing.
• Brokers spent vital time on administrative duties quite than specializing in high-value shopper interactions. Submit-call documentation, routine inquiries, and after-hours assist diverted dealer consideration from complicated shopper wants that required human experience and empathy.
• Lendi Group confronted the problem of scaling customized service throughout their intensive buyer base. Whereas their digital answer supplied comfort, sustaining the human contact that builds belief in monetary relationships proved tough at scale, particularly exterior enterprise hours.

These challenges led Lendi Group to discover how AI might remodel the mortgage expertise. Relatively than viewing AI as merely an effectivity device, Lendi envisioned a reinvention of the house mortgage journey—one the place expertise might anticipate buyer wants, present around-the-clock customized steerage, and free human specialists to give attention to constructing significant relationships.

Resolution overview

Lendi’s Guardian represents a basic shift in how prospects work together with their house loans. At its core, Guardian is designed to:

1. Monitor mortgage competitiveness by repeatedly scanning 1000’s of house loans each day and alerting prospects when higher offers change into out there
2. Observe fairness place in actual time as property values and trade circumstances change, giving prospects visibility into their present monetary standing
3. Streamline the refinancing course of with journeys that adapt to the shopper’s circumstances and auto populates varieties primarily based on inside and exterior knowledge sources, eradicating friction factors that beforehand deterred prospects from taking motion
4. Ship customized insights and suggestions primarily based on every buyer’s distinctive monetary state of affairs and targets

Lendi used Amazon Bedrock to speed up the construct of their agentic answer inside 16 weeks.

The answer is constructed upon Amazon Bedrock basis fashions and Amazon Bedrock Guardrails. Lendi selected Amazon Elastic Kubernetes Service (Amazon EKS) to deploy their AI brokers at scale, facilitating the mandatory infrastructure to fulfill shopper demand. By utilizing the wide selection of basis fashions (FMs) out there on Amazon Bedrock, Lendi was capable of choose task-appropriate fashions optimized for particular use instances.

A essential element of their answer is AI guardrails powered by Amazon Bedrock Guardrails, which assist make it possible for the shopper communications stay aligned with regulatory necessities. Moreover, Lendi developed Mannequin Context Protocol (MCP) servers to allow AI brokers to entry institutional information and work together with exterior providers seamlessly.

The important thing elements of the answer are as follows:

1. UI layer – Prospects work together with Guardian by an intuitive chat led interface built-in immediately into their Lendi dashboard, offering seamless entry to AI-powered mortgage insights and suggestions.
2. API layer – A RESTful API in Amazon API Gateway serves because the communication bridge between frontend purposes and backend AI brokers, dealing with request routing, authentication, and fee limiting to assist preserve safe and dependable interactions.
3. Compute layer – Amazon EKS hosts and orchestrates the AI brokers, offering auto-scaling capabilities to effectively deal with various buyer demand whereas sustaining constant efficiency and availability.
4. Intelligence layer – The core AI capabilities are powered by a number of specialised brokers constructed on Amazon Bedrock basis fashions. Lendi used Agno, an open-source agentic framework to develop these brokers, with MCP servers offering integrations to inside methods, exterior knowledge sources, and third-party providers. Bedrock Guardrails assist implement compliance boundaries, verifying that the shopper interactions adhere to Lendi’s communication pointers and stay centered on related mortgage-related matters.
5. Observability layer – Langfuse captures complete agent traces, together with inputs, outputs, reasoning chains, and efficiency metrics, offering full visibility into agent habits and enabling steady optimization and debugging. Amazon Cloudwatch logs are used to gather system degree logs.
6. Storage layer – MongoDB serves because the persistent knowledge retailer for consumer context, dialog historical past, and session state, enabling prospects to renew conversations throughout classes whereas offering brokers with the customer-specific context wanted for customized suggestions. Amazon S3 is used to retailer paperwork and information supplied by the shopper.

The next diagram illustrates the answer structure.

This structure sample gives a strong and scalable system to deploy AI brokers.

Agent circulation for mortgage refinance

Constructing upon this scalable structure, Lendi designed a multi-agent orchestration system the place specialised brokers can collaborate to finish the mortgage refinance journey. This modular method helps present a number of key benefits: clear separation of considerations between brokers, simplified improvement and upkeep of particular person agent capabilities, sooner response occasions by task-specific optimization, and simple troubleshooting when points come up.

The mortgage refinance course of flows by the next specialised brokers, with seamless handovers preserving full context at every transition:

• Mortgage Dealer Affiliate Agent (preliminary engagement) – This agent serves because the buyer’s first level of contact, embodying a pleasant, skilled persona just like a human mortgage dealer. Its main aim is to grasp the shopper’s present state of affairs and assess their curiosity in refinancing.
• Buyer Data Assortment Agent (knowledge gathering) – When a buyer expresses curiosity in refinancing, this specialised agent systematically collects important buyer particulars together with present mortgage info, employment standing, revenue, and refinancing preferences. The agent makes use of conversational methods to make knowledge assortment really feel pure quite than interrogative and gives clarifications to the shopper as required. The agent is context conscious and asks for info not already supplied by the shopper.
• Product Suggestion Agent (lender matching) – With full buyer info in hand, this agent analyzes the shopper’s profile in opposition to Lendi’s intensive database of lenders and merchandise. It presents appropriate choices with clear explanations of advantages and potential financial savings.
• Product-Particular Data Assortment Agent (utility preparation) – After the shopper selects their most popular product, this agent gathers the extra info required by that particular lender. Totally different lenders have various necessities, and this agent adapts its questions accordingly.
• Communication Agent (Linda) – Linda is the off-system engagement and re-engagement agent that retains prospects linked to their refinance journey, even after they’re not actively utilizing the Guardian system. Though the specialised brokers handle in-system duties from preliminary engagement to product choice and utility preparation, Linda operates throughout channels resembling SMS, electronic mail, WhatsApp, and push to carry prospects again in on the proper second. She detects when progress has stalled, surfaces well timed reminders or new alternatives, and reinvites prospects to proceed the place they left off. Drawing on stay knowledge from the Aurora Digital Twin, Linda tailors messages to the shopper’s particular context, tone, and aim, whether or not it’s encouraging them to reconnect their mortgage, assessment matched merchandise, or full their submission. In essence, Linda is the voice of Guardian past the app, serving to preserve prospects knowledgeable, motivated, and shifting ahead all through the refinance journey.

The next graphic illustrates this workflow.

This agentic method simplified the mortgage utility course of for patrons by offering an intuitive, pure language interface to share info, ask clarifying questions, and obtain steerage all through their refinance journey. For brokers, it alleviated the burden of guide kind filling and utility submission, liberating them to focus their experience on complicated buyer situations, relationship constructing, and offering strategic monetary recommendation the place human judgment and empathy are most useful.

Enterprise outcomes and buyer impression

Lendi’s Guardian utility is already delivering measurable outcomes, having settled thousands and thousands in house loans with refinance cycle occasions significantly sooner than Lendi Group’s baseline. Guardian extends this impression with its AI-powered Fee Radar, which scans 1000’s of house loans each day and permits refinancing in solely 10 minutes, with no paperwork, no cellphone calls, solely a single faucet. By automating routine monitoring and alerting prospects to raised charges in actual time, brokers can give attention to negotiation, empathy, and sophisticated structuring—the high-value, relationship-driven work that builds loyalty. Guardian launched in solely 16 weeks following a greater than 30,000-hour cross-functional dash, demonstrating how an AI-first structure accelerates each improvement velocity and buyer outcomes.

Classes realized

Lendi Group’s 16-week journey to construct and deploy the AI-powered Residence Mortgage Guardian supplied invaluable insights into implementing agentic AI at scale in a regulated monetary providers atmosphere. Listed here are the essential classes they realized:

• Prioritize early, iterative analysis metrics to information AI improvement systematically. Depend on data-driven metrics to make key choices resembling mannequin selection. Use Amazon Bedrock immediate administration for versioning prompts.
• Select fashions strategically by utilizing the varied mannequin choices of Amazon Bedrock. Acknowledge that essentially the most subtle mannequin isn’t at all times the best answer to your particular use case. Equally necessary is incorporating area information from human specialists into your prompts as a result of this contextual experience usually determines success greater than mannequin choice alone.
• Reap the benefits of utilizing Amazon Bedrock batch inference on duties that don’t require instant outcomes to cut back value.
• Deal with AI as a transformative expertise requiring daring imaginative and prescient and speedy, strategic implementation. Use the generative AI capabilities of Amazon Bedrock and the scalable cloud infrastructure of AWS to speed up AI-driven innovation.
• Prioritize accountable AI governance in regulated environments. Use Amazon Bedrock Guardrails to assist implement content material insurance policies, filter inappropriate responses, and preserve compliance alignment necessities all through the AI lifecycle.
• Stability automation with human experience. Design AI methods that increase—quite than change—human judgment, sustaining a customer-centric method the place human oversight stays central to essential choices.

Future roadmap

Lendi Group’s implementation of the AI-powered Residence Mortgage Guardian represents solely step one of their bold journey to change into a totally AI-based group by June 2026. With the inspiration now in place, Lendi Group goals to make use of agentic AI to rethink the entire mortgage and finance journey.

To assist this strategic initiative, Lendi is exploring new AWS providers, together with Amazon Bedrock AgentCore, which permits the deployment of brokers in a scalable and safe method with out the overhead of infrastructure administration. This method will additional assist speed up Lendi’s tempo of innovation.

“We’ve constructed our platform in order that refinancing occurs on the velocity of life, not on the velocity of paperwork,” says Devesh Maheshwari – CTO at Lendi. “A buyer can obtain a Fee Radar alert a couple of sharper fee or a shift in property worth throughout their morning commute. They faucet to have interaction with it and supply info to our Agentic platform “Guardian” and by the point they’re heading house, their refinance mortgage utility may be lodged. That’s not magic. It’s what occurs once you make investments correctly in clever automation, real-time decisioning APIs and a micro-services structure that coordinates every part from doc verification by to settlement, with out guide handoffs. The true problem wasn’t simply velocity. It was eradicating each level of friction whereas nonetheless assembly the best requirements of compliance and threat management. When your infrastructure can assist life-changing monetary choices in minutes quite than weeks, you’re not simply enhancing the expertise. You’re resetting what prospects anticipate from monetary providers.”

Conclusion

Lendi Group’s AI-powered Residence Mortgage Guardian represents a major leap ahead in how Australians handle their house loans. By utilizing the generative AI capabilities of Amazon Bedrock, Lendi has created an answer that helps remodel the mortgage expertise from a periodic, transaction-based interplay to an ongoing, proactive relationship that delivers steady worth to prospects. Wanting forward, Lendi’s journey to change into a totally AI-based group by June 2026 positions them on the forefront of innovation within the Australian mortgage trade. Their imaginative and prescient of AI built-in into “each workflow, each choice, each buyer expertise, and each dealer expertise” presents a basic reimagining of how mortgage providers may be delivered.

Concerning the authors

Deepak Dalakoti, PhD, is a Deep Studying Architect on the Generative AI Innovation Centre in Sydney, Australia. With experience in AI, he companions with shoppers to speed up their generative AI adoption by personalized, revolutionary options. Exterior the world of AI, he enjoys exploring new actions and experiences.

James Hardman James is a Senior Account Supervisor at AWS, partnering with Australia’s fintech and monetary providers organisations to navigate complicated expertise challenges. He works backwards from what issues most to his prospects, connecting them with the best funding, instruments, and specialist groups to assist them transfer sooner. James is especially centered on serving to prospects discover rising applied sciences like agentic AI – not for the sake of innovation, however to drive actual enterprise outcomes and higher serve their finish prospects.

Igor Londero Gentil is a Options Architect at AWS, primarily based in Sydney, serving to prospects design and construct on the cloud with a give attention to serverless and event-driven architectures. With a background spanning infrastructure engineering, cloud structure, and AI, he brings a practitioner’s perspective to fixing real-world issues — grounded in years of hands-on expertise earlier than becoming a member of AWS. Igor is a daily speaker on matters like event-driven architectures and AWS Lambda, and an energetic open-source contributor.

Devesh Maheshwari is the Chief Expertise Officer at Lendi Group Providers in Sydney, Australia, the place he’s driving the corporate’s transition to an AI-native enterprise. With greater than 18 years of expertise main expertise technique, digital transformation and engineering groups, Dev has a robust monitor document in fintech and extremely regulated sectors, shaping platforms that scale and ship actual enterprise worth. Earlier than Lendi and he has held senior management positions at DataMesh, Tyro Funds, Tabcorp & ThoughtWorks. He’s additionally a trusted advisor and mentor in tech, and he’s shared his insights on AI and innovation at trade occasions.

Samuel Casey started his profession within the startup ecosystem because the co-founder of a specialised AI consultancy. After efficiently spinning out a proprietary AI product and overseeing its acquisition by Mantel Group, Samuel joined Mantel 4 years in the past to guide high-stakes digital transformations. As an AI accomplice in Mantel, he has spearheaded a wide range of complicated initiatives for a broad vary of enterprise and authorities shoppers. Most just lately, Samuel has been on the forefront of the Generative/Agentic AI motion, devoted to serving to organisations combine AI Options into their core operations as these applied sciences have materialised within the international market.