All architectures
Proposed DesignBanking AIOpenStack / hybridAWSMLOps and CI/CD

Banking AI Copilot Platform — Agentic Credit Appraisal (Hybrid OpenStack + AWS)

A proposed target design. Components shown are planned or conceptual unless stated otherwise.

Full-stack target design for an enterprise banking copilot: SSO and RBAC, Flowise Agentflow V2 agents behind an MCP tool gateway, durable workflows, private LLM inference, GitLab MLOps and AWS staging.

Overview

  • Application layer: React/TypeScript analyst workspace, FastAPI + Pydantic backend, API gateway with OAuth2/OIDC single sign-on (Keycloak), RBAC, tenant isolation and audit logging.
  • AI orchestration: Flowise Agentflow V2 with credit-scoring, asset-appraisal, legal/policy and fraud/AML agents; an MCP tool gateway; private LLM inference (Llama / Mistral) on OpenStack; RAG with PostgreSQL + pgvector and Neo4j; connectors for Temenos core banking, ERP/CRM, REST, SFTP and email.
  • Workflow & execution: Temporal for durable, long-running workflows with human-approval waits, retries and audit history.
  • MLOps (GitLab) and AWS development/staging using de-identified data only.
  • Observability and operations: OpenTelemetry, Prometheus + Grafana, ELK/Loki, model and agent monitoring, backup/DR and cost monitoring.

OAuth2/OIDC, SCIM, multi-tenancy, tenant isolation and Temporal are proposed design elements of this target architecture.

Business use case

Target architecture for taking a credit-appraisal POC towards an enterprise-grade platform, with people approving every decision.

My role

Architecture design and documentation.