GenAI + ML Underwriting Copilot
A hybrid copilot that pairs structured ML risk scoring with LLM reasoning to draft underwriting recommendations directly from submission packets.
Cited underwriting memo
Output
End-to-end submission
Flow
Underwriter-in-loop
Review
Problem & use case
Underwriters spent the majority of submission time on data wrangling and narrative writing rather than risk judgment.
Solution & engineering approach
An XGBoost risk model surfaces structured insights; LangChain routes them through an LLM to produce a referenced underwriting memo inside a Streamlit workspace.
Architecture & data flow
- 01
Parse submission packet (docs + structured fields)
- 02
XGBoost scores risk + key drivers
- 03
LangChain assembles context for LLM
- 04
LLM drafts memo with citations to submission
- 05
Underwriter reviews + finalizes in Streamlit UI
Technology stack
Implementation status
Implemented — XGBoost + LangChain pipeline with Streamlit review interface