CustIQ 360°
Unified Customer 360 for relationship managers — minutes, not hours.
By
Swati FNU
Semester
Spring 2026
Problem
Relationship Managers in banks often need to navigate multiple core banking modules — CASA, Lending, Wealth, KYC — to understand a customer's complete financial profile, taking 15–30 minutes per lookup. This causes missed cross-sell opportunities, no real-time personalized recommendations, and difficulty answering what-if financial questions. Customer documents and images (ID proofs, salary slips, property documents) are still processed manually, slowing onboarding. Existing platforms like Finacle primarily manage structured data with limited AI capabilities.
Solution
A multi-agent AI system on top of simulated core banking modules that creates a unified Customer 360° profile. Specialized agents combine accounts data (Customer 360 Aggregator), let RMs ask natural-language questions (Conversational Query Engine), suggest next-best products (Cross-Sell Recommender), simulate scenarios (What-If Simulator), validate against compliance (Compliance Guardrail Agent), and push proactive alerts (Proactive Alert Engine). Goal: reduce lookup time from ~30 minutes to <2 minutes.
User flow
- RM opens the unified dashboard
- Asks a natural-language question about a customer
- Customer 360 Aggregator combines CASA, lending, wealth, and KYC data
- Cross-Sell Recommender suggests next-best products
- What-If Simulator instantly calculates EMIs, FD penalties, and loan scenarios
- Compliance Guardrail validates every recommendation
- Proactive Alert Engine pushes daily KYC expiry, FD maturity, churn-risk, and dormancy warnings
LLM components
- Information extraction — from documents and images (ID, salary slip, property documents)
- RAG and semantic search — over customer data and product catalog
- Conversational agent — for RM queries and mathematical reasoning
- Structured output generation — insights, recommendations, and rules-checked outputs
Tools
- Backend: Python + FastAPI
- Frontend: React
- Agent orchestration: LangChain / LangGraph
- Vector DB: FAISS
- LLM / multimodal: Gemini Pro & Vision API
- Vibe coding: AI assistants (Cursor / Claude / etc.)