Poject Aurora — Automating Customer Screening with Agentic AI

Client

Name: Project Aurora (name changed due to NDA)

Type: Neobank / FinTech Startup

Size: ~50 employees

Platform: Azure cloud infrastructure

Challenge

The client needed to automate their customer onboarding and compliance screening process. Manual screening by compliance officers was slowing down growth and increasing operational risk. As a regulated financial institution, they required a system with extremely high reliability — zero tolerance for LLM hallucinations or errors in compliance judgments.

Solution

Forma Pro designed and deployed an agentic AI-based solution leveraging Azure AI services, Azure Durable Functions, and Azure OCR to process and extract data from customer-submitted documents. A vector database stores structured customer profiles, while LLM-based agents orchestrate data analysis, risk checks, and decision support. A dedicated Copilot Studio workspace provides compliance officers with a human-in-the-loop interface, allowing them to review and confirm or override AI recommendations.

Reliability Strategy

Because customer screening must be error-free, the system integrates multiple safeguards: deterministic rules as guardrails, multi-agent cross-validation to reduce LLM hallucinations, and a human validation loop. Only consensus outputs are passed to compliance officers. This significantly improved accuracy and trustworthiness.

ArchitectureKey Takeaways

• Improved speed: onboarding time cut by 60%

• Higher accuracy: multi-agent and human loop eliminated LLM mistakes

• Better compliance oversight: Copilot Studio dashboard for officers

• Scalable design: durable functions and vector DB enable rapid growth