{
  "schema_version": "1.1",
  "id": "s8:https://www.youtube.com/watch?v=_gvamfT8H-w",
  "slug": "your-agreements-are-a-database-you-can-t-query-hiral-shah-docusign-sean-1aigd7h",
  "url": "https://feed7.dev/p/your-agreements-are-a-database-you-can-t-query-hiral-shah-docusign-sean-1aigd7h",
  "title": "Your Agreements Are a Database You Can't Query — Hiral Shah, Docusign & Sean Sodha, NVIDIA",
  "why_included": "Docusign and NVIDIA use a small extraction model to preserve complex agreement tables as structured data, showing where specialized parsers can beat generic VLMs in agent pipelines.",
  "summary": "Docusign says it processes **about 1 million agreements daily** across 1.9 million paying customers. Pricing, SLA, and rate-card tables often defeat line-oriented extraction, especially with merged or nested cells.",
  "practical_implication": "Use specialized extraction before retrieval when an agent must reason over table structure. The NeMo Retriever parse model preserves layout, reading order, and table relationships; Docusign reports **20× faster table extraction** than the open models it tested.",
  "agent_context": "Docusign says it processes **about 1 million agreements daily** across 1.9 million paying customers. Pricing, SLA, and rate-card tables often defeat line-oriented extraction, especially with merged or nested cells.\n\nUse specialized extraction before retrieval when an agent must reason over table structure. The NeMo Retriever parse model preserves layout, reading order, and table relationships; Docusign reports **20× faster table extraction** than the open models it tested.\n\nThe talk does not provide the underlying scores, model list, or production error rates. Docusign still uses a hybrid pipeline with OCR for other fields, while quantization and multi-token generation were described as future optimization paths.",
  "source": {
    "name": "AI Engineer",
    "url": "https://www.youtube.com/watch?v=_gvamfT8H-w",
    "published_at": "2026-09-16T15:30:28.000Z"
  },
  "source_class": "video",
  "content_type": "Video",
  "layer": "model",
  "domains": [
    "data"
  ],
  "topics": [
    "model-selection",
    "retrieval",
    "open-models"
  ],
  "verification": {
    "status": "source_linked",
    "label": "Source Linked",
    "method": "source_feed",
    "verified_at": null
  },
  "uncertainty": [
    "The talk does not provide the underlying scores, model list, or production error rates. Docusign still uses a hybrid pipeline with OCR for other fields, while quantization and multi-token generation were described as future optimization paths."
  ],
  "connected_context": {
    "meaning": "Retrieval cannot recover structure that ingestion has already flattened incorrectly. For agreements containing merged or nested tables, a layout-aware specialist parser becomes a prerequisite to dependable downstream reasoning, while the retained OCR path confirms that extraction should remain hybrid. The reported speedup supports task-specific model selection but lacks enough evaluation detail to establish accuracy or generality.",
    "corpus_size": 807,
    "generated_at": "2026-09-18T10:06:11.621Z",
    "connections": [
      {
        "title": "Teaching Nemotron Greek: Mining a Corpus, Adapting Retrieval, and Grounding Generation for Modern Greek across Specialist Domains",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2608.05138v1",
        "feed7_url": "https://feed7.dev/p/2608-05138v1-0bvu6le",
        "reason": "Both turn RAG into a staged, workload-specific selection problem: the target specializes parsing before retrieval, while the candidate adapts retrieval and grounding for a particular language and domain."
      },
      {
        "title": "When Does Bigger Help? A Controlled Study of LLM Scale for Ontology Learning",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2608.31118v1",
        "feed7_url": "https://feed7.dev/p/2608-31118v1-1abyqwd",
        "reason": "The specialized parser reinforces the finding that model scale alone is a weak selector; success depends on matching model architecture and evaluation to the structured task."
      },
      {
        "title": "Compression at the Edge — NVIDIA, Unsloth, HuggingFace, Ollama",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=J4_jCrTxMkk",
        "feed7_url": "https://feed7.dev/p/compression-at-the-edge-nvidia-unsloth-huggingface-ollama-0zq97jq",
        "reason": "Because quantization is only a proposed next step here, the candidate supplies the implementation consequence: any compressed parser artifact must be tested inside the real extraction pipeline for speed and quality."
      },
      {
        "title": "Local Models: Trust, Control, Optimization — Carter Abdallah, NVIDIA",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=FWMJQDH3iK0",
        "feed7_url": "https://feed7.dev/p/local-models-trust-control-optimization-carter-abdallah-nvidia-17u7gz9",
        "reason": "The parser is a concrete case for controllable open-model infrastructure, but its benchmark omissions also confirm that ownership and customization do not by themselves demonstrate task quality."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-09-16T15:30:28.000Z",
  "modified_at": "2026-09-16T15:30:28.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/your-agreements-are-a-database-you-can-t-query-hiral-shah-docusign-sean-1aigd7h",
    "json": "https://feed7.dev/p/your-agreements-are-a-database-you-can-t-query-hiral-shah-docusign-sean-1aigd7h.json",
    "markdown": "https://feed7.dev/p/your-agreements-are-a-database-you-can-t-query-hiral-shah-docusign-sean-1aigd7h.md"
  }
}