Qualifier V3 classroom prototype

Adaptive preliminary screening using encoded Indiana SNAP rules. This is not an official eligibility determination.

1 · Basic household
Do you live in Indiana?
How many people do you currently think are in your SNAP household?
Gross monthly household income ($)
Household composition checks

The tool uses these answers to warn when a person may be a mandatory household member rather than asking you to interpret the rule yourself.

2 · Age, disability & work rules
Your age
Paid work hours/week
Children you are responsible for

This prototype screens for obvious work-rule exceptions and flags cases where FSSA must verify an exemption or compliance.

3 · Student decision tree
Are you enrolled in an institution of higher education?
4 · Citizenship / immigration
Which best describes you for this screening?

Noncitizen SNAP eligibility changed in 2025 and depends on specific legal classifications. This prototype will not infer a classification from nationality or free text.

5 · Income & resources
Countable resources you can identify ($)
Income pattern

Grounded benefits assistant

Ask an Indiana SNAP question

Grounded AI mode. When deployed with the included server function, questions are matched to approved Indiana SNAP passages and only those passages (plus your screening trace when relevant) are sent to the AI. If the endpoint is unavailable, the tool automatically falls back to local grounded retrieval.

Ask general Indiana SNAP questions at any time. After screening, the assistant can also explain your specific result. If the approved material does not establish an answer, it will not guess.

Ask me things like “What is a countable resource?”, “Does my car count?”, “Who is in my SNAP household?”, “What deductions are allowed?”, “Can a college student qualify?”, or—after screening—“Why did I get this result?”
Approved sources in this prototype
Indiana FSSA — SNAP · Indiana FSSA — Income · Indiana FSSA — Policy Manual
Rules should be revalidated before production use because FSSA notes that its policy manual is continually revised.
Methodology & safety

How the prototype is designed

1 · Ground truthEligibility facts come from approved Indiana FSSA / government materials, not model memory.
2 · Rules engineThresholds and encoded eligibility branches are deterministic calculations.
3 · Grounded assistantThe assistant explains retrieved policy and the screening trace; it does not adjudicate eligibility.
4 · Fail closedMissing, conflicting, or unencoded rules produce review/verification—not a fabricated answer.

Production architecture: questionnaire → structured facts → verified rules engine → result + reason codes → retrieval from approved documents → LLM explanation with citations. The LLM never supplies thresholds or overrides the rules engine.

Qualifier · Deployable grounded-AI prototype · Official eligibility decisions are made by Indiana FSSA.