Working prototype for demonstration and staff testing. It is not presented as a production case-management, eligibility or automated decision system.
The problem
Workforce-program questions often begin with incomplete information and may require staff to move between program rules, agency resources, forms and referral sources. I wanted to explore whether a conversational interface could help staff reach the relevant program or source more efficiently while still asking for clarification when missing facts could change the answer.
What I designed into the workflow
The assistant can ask up to three clarifying questions when the answers could materially change the outcome.
Regression coverage includes OSY/YESS, SNAP, Ticket to Work and Dislocated Worker routing.
The workflow distinguishes AJC programs from external responsibilities such as Nebraska DHHS and SSA.
Testing checks that the form-pack workflow stays within the intended PDF-only rules.
Previous follow-up questions remain in history so short replies can still be interpreted in context.
The AJC logo is tested to return users to the Navigator home state.
Representative tested scenarios
The images below recreate representative test interactions aligned to behaviors documented in the V3.8 regression record. They are not production-user screenshots.
Testing discipline
The V3.8 test record documents successful checks for the structured follow-up schema, follow-up limits, prevention of unnecessary or repeated questions, PII-aware clarification rules, conversation-history preservation and existing answer/form-routing behavior. Regression checks also cover OSY/YESS, SNAP → Nebraska DHHS, Ticket to Work → SSA, Dislocated Worker layoff routing, PDF-only form-pack safeguards and AJC-logo navigation.
follow-up questions per assistant turn
named program-routing scenarios represented in the regression record
PDF-only form-pack safeguard regression check
follow-up history and clarification UI checks
What this project demonstrates
The project goes beyond building a chatbot interface. It demonstrates requirements definition, iterative AI-assisted development, workforce-program routing logic, conversation design, UX refinement, safeguards and regression testing — translating a complicated staff knowledge problem into a testable prototype.
Scope
The Knowledge Navigator is a prototype intended for demonstration and staff testing. I am not presenting it as an approved production system or as evidence of measured staff or participant outcomes.