What it is
Keep the review question in code; let hosted Jev answer six atomic checks per abstract.
What it does: Builds a typed System One call for each PubMed record, ranks uncertain items for humans, and supports mock or live modes.
Why it belongs in the showcase: Clear hosted TypeSafe SDK usage on life-science abstract triage, complementary to abstract-sift.
Caveats: Single-question demo; verify against current SDK docs.
Evidence
Public MIT repo, created 2026-10-30, 0 stars as of 2026-10-30. README cites docs.typesafe.ai and typesafe-sdk 0.7.4, documents mock triage/evaluate without a key, and live triage with TYPESAFE_API_KEY and model jev-1.13.0. Distinct from ValiusSciences/abstract-sift (general interest screening vs this fixed oncology/FMT question demo).
Caveats
- Demo is scoped to one published review question, not a general screener product
- Live runs need TYPESAFE_API_KEY and bill TypeSafe usage
- Screening accuracy is not claimed beyond the repo's own evaluate helpers
Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.