CaseList

Inspiration

Counselors need inspectable evidence behind a college list.

What it does

CaseList extracts exact experiences, compares confirmed facts against fictional applicant records, and builds a provisional six-college list. Counselors inspect cases, uncertainty and corrections.

How we built it

Python, HTML and JavaScript with 60 synthetic applicants and 140 applications. Deterministic code assigns categories. Nicrron routes Claude Haiku 4.5 and GPT-4o mini through independent opinions, anonymous peer review and chairman synthesis inspired by Karpathy’s LLM Council. Source validation preserves exact citations and policy categories. PRISM supplies input Guardrails, Knowledge Base grounding, live traces and classified-output evaluation.

Challenges and learning

Model agreement can contain wrong citations. We validate sources and fail closed. Shared quotation references keep evaluation within bounded context.

Limitations and next steps

All records and colleges are synthetic. Categories are provisional, not validated admission predictions. Live verification is per run. Representative data and counselor validation are needed before real use. App runs locally; video and judge access remain pending.

Built With

• python