The Fast and the Curious: Harnessing AI in Financial Data Extraction at Scale
summary
A production pipeline scrapes millions of scanned PDF filings, classifies pages, extracts 38 million fields across 3.6 million pages and validates results with arithmetic checks plus a tiered human-refinement workflow, cutting extraction time by 40x speed increase and driving compute plus human costs to 0.068 EUR per report (compute alone ~0.04 EUR). Built in one year by a two-developer team using Python, Laravel, OpenAI, VoyageAI, Azure AI Document Intelligence, ChromaDB, the system yields 86% flawless extractions, 14% flagged for review, and already supplies commercial customers.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:38Nikolas background
- 3:41How the website looked
- 4:12I was expecting no
- 4:35Jack Daniels
- 6:19Overview
- 6:52The Problem
- 7:53Public Registry
- 10:24Income Statement
- 12:32Implementation
- 14:30How it works
- 16:59Data refinement system
- 18:32Expert level verification
- 20:22Cost efficiency
- 22:08Cost per report
- 23:04Commercialization
- 23:46Framework
- 26:07Summary
- 26:51Questions
related talks
32mAutomate data extraction & analysis with intelligent document processing
30mAI and API, made for each other
22mStructuring the Unstructured : Empowering the AI Workflows
50mAzure AI, Power new possibilities for every organization
20mDash of AI: Intelligent Automation to eliminate repetitive tasks
31mAI for Everyone: Unleashing the Power of Open Source Tools
32mBeyond Bolt-On AI: A New Way to Build Software
18mSmall AI Teams with Huge Impact