Interpretation engine
The drafting prompt is versioned product code that encodes AMP/ASCO/CAP tiering and disease-specific frameworks. It classifies each variant independently, surfaces evidence conflicts, and refuses to invent therapy claims without tier-I support.
Tier I–IV with explicit oncogenicity, never blurred across variants. Insufficient evidence becomes a documented VUS.
ELN 2022 and 2024 scored side by side; flags when the update changes the outcome. IPSS-M for MDS.
Detects patterns that modify risk beyond single variants — NPM1+FLT3+DNMT3A triplet, biallelic TP53, spliceosome signatures.
Auto-detects DDX41, RUNX1, GATA2, CEBPA and other WHO-recognized genes; routes them to ACMG germline review, not somatic tiers.
Knowledge suite
Curated hemato-onc annotations from OncoKB, ELN 2022, WHO 2022 and NCCN — plus the calculators, wizards and workbenches pathologists reach for mid-case. Versioned, with a 30-day changelog.
ELN 2022 · IPSS-M scoring with live inputs.
DDX41 · RUNX1 · GATA2 predisposition workup.
Weigh evidence for variants of uncertain significance.
Tiering rules from Li et al. 2017, searchable.
Retrieval over guidelines and the lab's own notes.
FLT3 · IDH · TP53 trial eligibility surfacing.
Pharmacogenomic dosing context at a glance.
Reusable note blocks and report scaffolds.
Drill AMP tiering and ELN criteria.
Lab intelligence
Beyond the single case: longitudinal patient tracking, cohort analytics, and surveillance that watches for knowledge-base changes affecting cases you've already signed out.
Serial biopsies per patient, with emergent and clearing clones surfaced automatically.
Mutation trends, co-occurrence heatmaps, and panel performance across your cohort.
Dedicated trackers for clonal hematopoiesis and measurable residual disease over time.
When OncoKB or a guideline changes, flags signed-out cases whose interpretation may now differ.
Track where pathologists agree with the AI draft and where they override — by gene and over time.
Workflow & collaboration
Smart Queue ranks the backlog by urgency. Batch operations, routing rules, tumor board and external second-opinion sharing move cases through the team.
Priority-sorted by risk, STAT flags and TAT — urgent cases never sink.
Assemble MTB packets; share a read-only case link with an external reviewer.
Re-run a cohort against an updated prompt; sign out a batch in one reviewed action.
Compliance & audit
Every AI call, edit, amendment and sign-out is logged with model, prompt version and tokens — built for CAP/CLIA review, not bolted on after.
Concordance, amendment rate and QC indicators tracked continuously.
Full request payload, raw response, parsed output and token counts per call.
Signed reports are frozen; amendments create a linked new version with a reason.
Disease coverage
AML is the deepest, with ELN 2022/2024 risk and composite-mutation logic. Each context applies the right classifier and the right driver set.
Validation
HemeCopilot's interpretation prompt is versioned in git and changed through review. An expert-validated eval set runs on every change — tier accuracy, risk accuracy, therapy recall, germline flags and uncertainty quality. A regression over the threshold blocks the merge.