Implementing AI Document Processing at Scale: Syngenta

on YouTube ↗43m

" summary

Syngenta builds Wingman invoice pipeline that processes 1 million invoices (2.5M related docs) per year, extracts fields with OCR+LLMs and stores results in an invoice data lake (Redshift, S3, OpenSearch Serverless) fed by Glue and EC2-hosted SAP. Proof-of-concept runs tax checks for Argentina (4,000 invoices/year) and flags mismatches in PowerBI; it works in batch and is moving to desktop chatbot and near‑real‑time checks. Core tooling: Amazon Bedrock Data Automation (BDA), Bedrock models (Entropic, Nova experiments), Textract-like OCR, and prompt-tuned blueprints.

AI-written summary · about

" chapters (6)

1. 0:00Introduction and overview
2. 4:58What is intelligent document processing
3. 10:00Syngenta and AI overview
4. 21:03Wingman project deep dive
5. 30:18Amazon BDA demo and features
6. 38:54Wrap-up and lessons learned

" related talks

• 30mIntelligent document processing: Extract data & insights from documentsSandeep Mistry · AWS re:Invent 2020
• 32mAutomate data extraction & analysis with intelligent document processingAshish Khare, Saumin Patel · AWS Summit Online – Americas 2021
• 21mRealizing the Potential of Business AIFeiyu Xu · hub.berlin 2022
• 32mAutomate data extraction & analysis with intelligent document processingAshish Khare, Saumin Patel · AWS AMER Summit May 2021
• 30mIntelligent document processing for the insurance industryRaj Copparapu, Reddi Gudla · AWS re:Invent 2020
• 32mUsing AI to power a circular economy for aerospaceClaudio Pontili, Matthieu RITTER, Tom METZELER · AWS re:Invent 2023
• 35mAutomate your document pipeline with intelligent document processingPrem Ranga · AWS Machine Learning Summit 2021
• 38mFrom Paperwork to Pipeline: Autonomous Mortgage AINaganarendar Chitturi · Conf42 MLOps 2025