AI Engineer World's Fair 2026 — unofficial import demo

Agreements power every enterprise business, but the most critical data — pricing schedules, SLA obligations, rate cards — is often trapped in tables that traditional extraction tools destroy.

This session shows what changes when you can actually extract that data accurately at scale and make it searchable.

We'll walk through the before and after:
Before: Contract tables require manual review. Rate cards are buried. SLA terms are scattered across exhibits. Procurement teams spend hours piecing together pricing structures — and searching for specific terms means opening every document.
After: Tables are automatically extracted, structured, and queryable. Operations teams can surface SLA notification requirements on demand. Legal can answer "what hourly rate did we agree to?" in seconds.

Docusign will share what we've achieved evaluating NVIDIA Nemotron Parse for our document processing pipeline, including how we tested against real enterprise contracts (not synthetic benchmarks), why we're serving the model via vLLM, and what it takes to turn extracted table data into searchable, retrievable agreement intelligence.

NVIDIA will cover the architecture behind Nemotron Parse and where the model is heading — including how NeMo Retriever's embedding and reranking models connect extracted data to search and RAG-based applications.

Attendees will leave with a realistic view of where vision-language models excel at document understanding, where the gaps remain, and how to think about building searchable contract intelligence into their own systems.