Regulatory Pricing Records, From a Scanned Backlog to Structured Data

Pricing notifications, circulars, and regulatory filings for a large medicine portfolio accumulate for years — scanned PDFs, tables, multiple formats, some in regional languages — each one carrying a medicine name, manufacturer, MRP, ceiling price, effective date, and a notification reference. Extracting that by hand, filing by filing, across years of archive doesn't scale, and new filings keep arriving every month regardless of how far behind the backlog is.

The harder part isn't reading any single filing — it's knowing whether a new filing is a brand-new price, a revision of an existing one, or a duplicate of something already on file. That judgment call requires context from everything filed before it, not just the document in front of you.

The Saransh Document Intelligence Agent classifies and extracts structured fields — medicine name, manufacturer, MRP, ceiling price, effective date, notification reference — from scanned and tabular filings, regardless of format or language. A Saarthi orchestration layer then chains the rest of the pipeline: cross-validating each new extraction against the existing record set to determine whether it's a new price, a revision, or a duplicate, and flagging genuine anomalies for human review.

That cross-validation step is why this needs to be agentic rather than a one-time extraction script. Processing a new filing correctly means holding the context of years of prior filings and making a judgment call about how this one relates to what's already on file — not running the same extraction logic in isolation on every document that arrives.

Reviewers work from structured, cited records instead of re-reading each filing from scratch, and the extraction pipeline keeps pace with new filings each month rather than needing to be re-run as a one-off digitization project.