Panchkula houses many Haryana government offices, and residents deal with state e-services for certificates, pensions, scholarships and welfare schemes. Citizen-service assistants that answer “Am I eligible?” are an obvious GenAI use, and an obvious risk: eligibility depends on exact combinations of age, income, category, district and documents, which language models paraphrase loosely.
Read moreRead less
Testers already know the right tool for this: the decision table. Applied to an AI assistant:
- Write the rules of an invented scheme as a short public-style notice, then derive a table of conditions and outcomes from it.
- Turn every row, plus the boundary values (age exactly at the limit, income one rupee over), into a conversational question.
- Pass only answers that match the table and quote the clause; fail any answer that sounds encouraging but is wrong.
- Check that the bot recommends confirming with the issuing office and never claims a final decision.
- Repeat the same rows in Hindi and Punjabi.
Boundary values are where models fail most, so this exercise usually finds real defects quickly. It also shows interviewers that you bring classic test design to AI rather than treating it as magic.