Proteron
Enterprise AI CRM for the Insurance Industry
An enterprise CRM for the insurance industry, built at Astheron Technologies.
Case Study
The problem
Insurance agents lose time hunting a single clause across hundreds of pages of policy documents, and underwriting teams verify claim files by hand, one at a time. Both are reading work — and both slow down closing a policy.
Architecture
Two runtimes with different responsibilities. NestJS (TypeScript) owns CRM operational logic — pipeline, leads, policies — where transactional consistency decides correctness. FastAPI (Python) owns AI document computation, where the library ecosystem is far more mature. PostgreSQL with pgvector serves semantic policy search, and Redis sits between the two as the cache layer.
The hard part
Splitting two runtimes only pays off if the boundary is drawn in the right place. AI document computation is long-running and hard to predict, while CRM operations must stay responsive — running both in one process means a single heavy claim file can stall the entire agent pipeline.
Outcome
Agents find policy articles, exclusion clauses and coverage comparisons in seconds through a RAG-based agent, while medical claim and identity verification runs automatically through OCR and VLM — with that heavy work isolated from day-to-day CRM operations.
System Architecture
Runtime
NestJS · TypeScript
CRM operational logic — pipeline, leads, policies — where transactional consistency decides.
FastAPI · PythonAI
AI document computation, where the library ecosystem is far more mature.
Data
PostgreSQL · pgvector
Serves semantic policy search.
Redis
Cache layer between the two runtimes.
Why it is shaped this way
Two runtimes with different responsibilities, each chosen for where its ecosystem is strongest.
Engineering Decisions
Decoupled Microservice Architecture
Separates enterprise CRM operational logic in NestJS (TypeScript) from heavy AI document computation in a FastAPI (Python) microservice.
Policy Document Copilot
A RAG-based AI agent helping insurance agents find policy articles, exclusion clauses and coverage comparisons in seconds.
Underwriting & Claim Verification
Intelligent OCR and VLM modules that automatically verify the validity of medical claims and policyholder identity documents.