An Agent That Counts My Receipts, Not My Claims

Developer Keniel Zepeda has released a specialized AI agent designed to query Sanity-based records with high precision, prioritizing verifiable evidence over raw claims. Built for the Sanity Challenge, the project addresses the common issue of AI hallucination by enforcing strict retrieval rules and output contracts. The agent uses two distinct instruments: a Knowledge Base for prose-based queries and GROQ for structured data lookups. By maintaining separate fields for status and expiry, and requiring explicit citations for every answer, the system ensures that it only reports what can be factually proven. The project includes a public demo, a CLI harness, and a transparent evidence trail of its development and testing phases. This approach demonstrates a robust method for building self-correcting systems that maintain data integrity and provide verifiable, citation-backed responses, rather than attempting to infer certainty from incomplete or ambiguous datasets.