Sourcepoint, a Firstsource company and leading provider of outsourced mortgage services and solutions, announced its investment in building its own “domain-centric” Large Language Model (LLM) specific to the mortgage process. This investment leverages the company’s domain expertise to make sector-specific, AI-driven services and platform offerings.
This initiative follows the launch of Firstsource relAI, an AI-focused digital transformation suite of platforms, solutions and offerings. It provides clients with more efficient operations, improved decision-making and personalized experiences by leveraging natural language processing, LLMs and advanced statistical analysis to provide intelligent and effective search capabilities across various content types and sources.
Sourcepoint’s LLM reduces the cycle time for pre-qualification and formal loan applications. This creates a digital end-to-end process for loan application and fulfillment that balances the convenience of self-service with the personalized support of a loan officer. It will be equipped with capabilities like classification, extraction, recommendation and summarization. These capabilities will support credit assessments, loan quality checks, mortgage document digitization and risk assessment, including insights into loan receivables management and process automation. The LLM will also accelerate the development of AI co-pilots and autonomous workflows.
“This initiative aligns with our ‘inch wide, mile deep’ philosophy of bringing our deep domain expertise to our customers through deep-tech intervention,” Ritesh Idnani, managing director and CEO at Firstsource, said. “Generative AI is transforming both our business and our clients’ operations. … Mortgage is a critical domain for us, and we are uniquely positioned to capture the opportunity with the interplay of domain knowledge and technology.”
“It’s important that our deep domain expertise in mortgage is embedded in this new Gen AI model,” Hasit Trivedi, chief digital and AI officer at Firstsource, added. “This ensures that knowledge is democratized and widely applicable across our engagements. We are using a unique approach by combining transformer architecture with techniques like model blending.”