The report documents 4 LLM experiments translating SNAP and Medicaid policy into executable rules across 7 states, validating the feasibility of GiveCare's rules-as-code approach to benefits eligibility.
Experiments revealed that LLMs can parse complex conditional logic in policy documents but require structured validation layers to catch edge cases, informing GiveCare's neuro-symbolic verification design.
Cross-state variation in policy interpretation was a primary challenge, supporting GiveCare's state-specific eligibility configuration architecture.
The Beeck Center recommends pairing LLM translation with human policy-expert review, reflected in GiveCare's human-in-the-loop validation workflow for eligibility rules.
Findings demonstrate that rules-as-code reduces eligibility determination errors compared to caseworker-only processes, strengthening the case for GiveCare's automated screening.