Agentic AI portfolio demo by Chantell Harris-Headley. This interactive demonstration shows how a Supervisor coordinates specialized Credit, Income, Asset, Collateral, Critic, and Decision agents across a structured underwriting workflow.
Sequential routing and shared workflow state across multiple specialist agents.
DTI and LTV are calculated outside the language model for reproducibility and auditability.
A dedicated Critic layer reviews specialist outputs, flags risk, and determines escalation needs.
Borderline and high-risk cases are explicitly marked for human review rather than fully automated disposition.
The source project demonstrated PII masking before data is provided to an LLM.
The original project used ChromaDB-based retrieval for mortgage policy context.
Python • LangGraph • LangChain • OpenAI models • ChromaDB • RAG • deterministic financial tools • multi-agent orchestration • critic review • human-review escalation
Full technical repository: GitHub