AI CFO India: A Deterministic-First, Retrieval-Grounded Multi-Agent Architecture for Explainable, Compliance-Safe Portfolio Intelligence for Indian Retail Investors

Abstract

This paper presents AI CFO India, a deterministic-first, retrieval-grounded multi-agent architecture for explainable and compliance-safe portfolio intelligence for Indian retail investors. The system separates deterministic financial computation from AI-generated explanation. Seven independent deterministic agents compute portfolio findings, while a single isolated LLM-based Explainability and Synthesis Agent provides grounded narration. A retrieval-based knowledge base and hard compliance engine are used to reduce hallucination and prevent directive or advisory language. The implementation uses FastAPI, PostgreSQL, pgvector, and a React/TypeScript frontend. An implementation-level evaluation reports 65 passing automated tests covering unit, integration, resilience, and compliance behaviour.