Banking AI Copilot — Credit & Asset Appraisal
Private credit and asset appraisal POC with human decision authority
Rackspace Technology · ACB Bank
Private banking analyst POC — not a live bank production deployment. The public demo is a separate showcase environment.
A private credit and asset appraisal proof of concept that uses specialist agents, document RAG and cited evidence to streamline reviews — while human decision-makers stay in control.
Proposed target architecture — not the POC as built
Problem & business context
Credit and asset appraisal reviews involve many documents, policies and specialist checks. The bank wanted to explore how AI could speed up reviews without giving up privacy, compliance or human decision authority.
My role & contribution
Led the workflow, data and privacy discovery workshops and built the POC hands-on.
What I built or designed
- FastAPI customer, document and audit API services, plus a banking analyst workspace for intake, evidence retrieval and staged human review
- Flowise Chatflow orchestration for policy and citations, model routing and structured responses
- Separate document RAG with Ollama embeddings and PostgreSQL/pgvector
- Restricted Temenos / agent “digital officer” patterns for credit, asset, legal and fraud review — human officers keep final authorization
Architecture & workflow
The reference architecture was extended with OAuth/OIDC/SCIM, tenant-aware RBAC, Temporal and GitLab + SageMaker MLOps. These are planned extensions, not verified deployed features. The related diagrams show these proposed target designs.