Agentic Intent Capture: AI Agent on Bedrock

The digital mortgage lender — the consumer mortgage business inside the parent company — is the U.S. online-mortgage leader, with trillions in originations since inception and recent acquisitions in real estate search and mortgage servicing. The technology platform is lender-built and lender-operated, AWS-heavy: SageMaker AI/Studio, EMR, S3, Athena, Lake Formation, Glue, Redshift, Lambda, EKS plus Istio, MLflow, model registry, feature store. Taller's engagement at the digital mortgage lender is the newest in the portfolio: active 2025–2026, sourced primarily from Argentina and Brazil, and anchored on individual contributor roles across Frontend, Backend, Cloud Infrastructure, Salesforce, and legacy modernization. Three cases below describe what Taller's engineering capacity supports at the lender.

The lender's AI Agent strategy is a centralized AI Agent API routing requests to eight domain-specific Bedrock Agents across origination, servicing, and broker support. Each agent has its own knowledge base, memory, action groups, and guardrails, with return-of-control to human bankers for decisions a regulated lender cannot delegate. The architecture is lender-built; the platform requires AI-relevant engineering capacity to operate at scale within the lender's regulated-lending guardrails (IAM, KMS, STRIDE and SHOSTACK threat modeling, OWASP Top 10 for LLM Applications).

Taller's same-client engagement at the lender includes AI-relevant individual contributor roles — the strongest AI signal across the Vol I client base — plus Cloud Infrastructure Engineering aligned with the AWS, EKS, and Bedrock stack the lender operates. The capability reference Taller carries on production agentic AI in regulated environments:

The multi-agent generative-AI framework on AWS Bedrock with LangGraph, LangChain, knowledge graphs, RAGAS evaluation, and the productivity AI deployment with hybrid search, multi-query retrieval, and React with WebSocket streaming.

Same-client AI-relevant individual contributor presence at the lender across the AWS Cloud Infrastructure substrate the lender AI Agent depends on.

The lender's AI Agent outcomes speak for themselves: visitors three times more likely to close a loan after engaging the AI Agent, an eighty-five percent decrease in transfers to customer care, a forty-five percent decrease in transfers to servicing specialists, sixty-eight percent high satisfaction in chat, and an expected thirty-three percent improvement in lead conversion. These are the lender's outcomes earned by the lender's eight-Bedrock-Agent architecture, the knowledge-base and action-group design the lender teams own, and the guardrail discipline (STRIDE, OWASP Top 10 for LLM, return-of-control).