MoodSpan
Clinical search, safety, and evaluation.
A focused packet for reviewers: what the product does, how answers are grounded, where quality is measured, and what remains blocked.
438
Gold retrieval queries
Bootstrap intervals tracked
92.0%
Recall@5
Internal same-corpus retrieval
25
Screeners
Validated instruments and scoring
6
Comparison guides
High-confusion clinical searches
Operating model
Four controls make the demo worth showing.
The system is strongest when it shows both capability and constraint. The brief keeps those two facts visible at the same time.
Evidence Retrieval
Queries are rewritten into clinical search intent, matched through hybrid retrieval, and ranked before an answer is drafted.
Grounded Response Contract
The response path checks retrieval confidence, citation coverage, and unsupported clinical claims before the answer is released.
Founder Observability
Admin surfaces track weak traces, groundedness failures, low-confidence retrieval, and release-risk buckets.
Public Safety Boundary
Every clinical surface keeps education separate from diagnosis, treatment decisions, and emergency care.
Demo sequence
A short path through the product without sales copy.
- 1
Start with the brief to establish scope and quality controls.
- 2
Open the live runner to show streaming events, retrieval metadata, citations, and response boundaries.
- 3
Use a hard differential query, then inspect the sources and caveats.
- 4
Finish on the research page for metrics, architecture, and benchmark limitations.
Ready for review, not final medical authority.
The useful story is the engineering discipline: sources, uncertainty, quality gates, and clear limits.