ai-lens
Healthcare AI evaluation, visually. Capability benchmarks are saturated; the evaluation that answers
"does this help?" looks like product analytics — and it happens in private.
deckState of Healthcare AI Benchmarks15 slides in four parts — the record · the ceiling · behind the wall · the instrument
storyOne chart, then the argumentThree acts — the era timeline assembled and recolored, the ticket that never closes, and where the compute actually belongs
one-pagerOne recommendation, walked all the wayARISE's turn-level gates and the ticket's loop on a single line, ending at the outcome — with a scorecard of our effort
webgpuThe lens canvas — GPU editionJelly blobs, five scenes, draggable cards with the papers inside, TUI leaderboard (TypeGPU)
htmlThe lens quadrant — DOM editionSame 18 artifacts and lenses, no WebGPU required
one-pagerWhat to monitor for an embedded AI assistantSix layers on the visit clock; where public benchmarks sit on the same axis
3d tiltThe chart has a basementThe era gantt tilts back like a lifted floorboard to reveal the private funnel beneath — particle rain drains each era into its basement bin
field guidePhysician prompting techniquesHow clinicians actually drive grounded medical AI — adjudication, output contracts, admin offload, tool-chaining
essayModel evaluation should look like product analyticsThe long-form argument, with references