Developer Builds AI Agent That Verifies Receipt Counts Against a Sanity CMS Dataset
A developer has built and deployed an AI agent called 'ask-the-record' that queries a structured Sanity CMS dataset of 120 documents — including articles, findings, people, and claims — to verify factual assertions rather than accept them at face value. When asked how many receipts support a finding attributed to three engineers, the agent returned one verified receipt instead of three, because only one was traceable in the comment records. The agent is publicly accessible at ask-the-record.vercel.app without a login, offering five preset questions with measured response times ranging from 13.8 to 33.8 seconds. Each answer includes a verdict drawn from a fixed set of statuses, a list of sources read, an evidence date from the record rather than the current date, and a mandatory uncertainty field that is never left blank. The project, submitted for the Sanity Challenge, separates claim status and expiry status at the schema level and uses a boolean patch field to report merge status without inferring outcomes beyond what the data contains.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in