Skip to main content
Download free report
SoftBlues
SoftBlues
Back to Projects

AI-Powered Loan Application Processing Platform

Solution Architect2025Oleksandr S.
OS
Oleksandr S.

AI / Cloud Engineer

LLM & AI Agents

Key Expertise

Agentic AI SystemsSolution ArchitectureAI-Powered SDLCRAG & Vector Search

Experience

12+ years

Timezone

CET (UTC +1)

Skills

AI / ML

LangChain4jClaude CodeGoogle GeminiOpenAIAWS Bedrock

Languages

Java

Databases

MSSQLCockroachDBCassandraWeaviateHazelcastAuroraPostgreSQLMySQLDynamoDB

Infrastructure

MavenMSKAWS LambdaTerraformSNSJenkins CICloudWatchSQSS3DockerECS/Fargate conceptsApiGeeGradleGlueApache ShiroDatadogAWS Cognito

Frameworks

AWS SageMakerSpring BootPySparkSpringAICursor

Integrations & Protocols

OcrolusNewRelicKoogFeignSumoLogicApache KafkaMCPLoanPro
7-day risk-free trial
Response within 24 hours
View Full Profile

Overview

Architected and developed a sophisticated loan application processing platform with agentic AI capabilities for a fintech lending company. The system combines an N8N-inspired declarative workflow engine with LLM-powered agents equipped with 30+ MCP tools for autonomous task execution. Core modules include automated bank statement extraction and fraud detection, vector-powered semantic search for document analysis, and multi-tenant architecture with complete data isolation across 76 database tables.

Achievements

The platform automated complex loan underwriting workflows, significantly reducing manual review time through configurable knockout and auto-approval thresholds. The agentic AI system autonomously handles multi-step research and verification tasks using 30+ MCP tools spanning browser automation, financial processing, and web search. A 2-pass LLM transaction classification pipeline with confidence scoring improved categorization accuracy. Multi-provider LLM architecture enables seamless switching between five providers without code changes.

Responsibilities

  • Defined end-to-end system architecture: designed the workflow engine model, agentic orchestration layer, financial processing pipeline, and vector search infrastructure as cohesive modules with clear contract boundaries.
  • Architected an N8N-inspired workflow engine with 15+ node types, declarative JSON-based workflow definitions, and configurable execution states.
  • Designed and implemented agentic AI orchestration layer with 30+ MCP (Model Context Protocol) tools covering browser automation, business data retrieval, financial processing, and web search/deep research capabilities.
  • Designed financial processing module: bank statement PDF extraction, 2-pass LLM transaction classification with confidence scoring, automated fraud detection, and configurable knockout/manual review decision logic.
  • Architected vector search infrastructure with RAG pipeline, defined embedding and chunking strategies for semantic document analysis and retrieval using Weaviate.
  • Designed multi-tenant data architecture with JSONB-based flexible entity storage (76 tables), event- driven audit system, and webhook notification pipeline.
  • Built comprehensive observability layer: LLM metrics tracking, workflow audit trails, structured logging, and dynamic configuration management.

Technologies Used

JavaSpring BootPostgreSQLWeaviateGoogle GeminiOpenAIAWS BedrockMCPMavenAWS CognitoFeignOcrolusLoanPro
OS

This project was delivered by

Oleksandr S.

View Full Profile

Ready to Build Your AI Team?

Get matched with the right AI experts for your project. Book a free discovery call to discuss your requirements.

We respond within 24 hours.