All workFILE 03 / 16

Wealth management

Millions of securities, one grounded answer

Shipped for A wealth-management fintech · led by Nikhil Bery

01

The problem

Advisor and client questions spanned four separate market-data providers, so a grounded answer meant manual lookups across all of them.

02

The constraint

Every answer had to stay grounded across four market-data providers at once, not stitched together by hand from separate lookups.

03

The system

A multi-agent financial assistant over four market-data sources, a PDF extraction pipeline for 50+ page statements, and hybrid search across millions of securities, behind a 40+ endpoint backend on serverless microservices.

FIG. 01 · SYSTEM FLOW · 7 NODES · 6 FLOWS

50+ page statements → PDF extraction pipeline (95%+). 4 market-data sources → multi-agent assistant · 4 frameworks. hybrid search · millions of securities → multi-agent assistant · 4 frameworks. PDF extraction pipeline → multi-agent assistant · 4 frameworks. multi-agent assistant · 4 frameworks → grounded answer. grounded answer → 40+ endpoint backend

04

The outcome

95%+ extraction accuracy on 50+ page statements at 10 to 30 positions per second, with four agent frameworks unified behind one assistant.