A clinical decision companion for practicing oncologists — bringing the right answer within reach of every patient.
The same patient, different prescriptions — depending on the oncologist they meet.
The gap is not in competence. It is the structural limit of time, knowledge, and memory.
One widely publicized oncology AI achieved 87.9% concordance with experts¹ — and still failed. Failure didn't come from the technology. It came from the asymmetry of trust.
Every clinical claim links to a primary paper or authoritative guideline. The clinician verifies in one click — never accepts on faith.
Plausible-sounding answers never appear. The system explicitly states “insufficient evidence” — freeing the clinician from hallucination risk.
If a mutation isn’t in the clinician’s input, the system never assumes it. Only the clinician’s own facts ground every output.
Output is always presented as options and considerations. The final judgment remains the clinician’s — authority is never displaced.
History, symptoms, labs, imaging, molecular results — in plain prose. The system extracts the structure automatically.
“62 y/o male, right hemicolectomy 3 yrs ago. Recent CT shows multiple liver lesions (S6, S8), suspicious for mets. CEA 73.9 ng/mL (rising). KRAS G12C positive, MSS, HER2 negative. ECOG 1, eGFR 75. Progression after 1L — need 2L options.”
35+ clinical parameters — every variable the case warrants:
Most medical AI answers once and forgets. OncoSignal accumulates, remembers, and adapts.
Each clinician’s patient records are accessible only to them. Identifiers are scoped to the account.
Every analysis for one patient stacks chronologically: diagnosis, 1L decision, follow-up scans, RECIST response, progression, 2L decision.
Each new lab, scan, or RECIST result updates the recommendation. Next-line candidates, dose adjustments, matching trials all refresh on history.
“What did I see for this patient last year?” One patient ID unfolds the full history. The system supplements clinical memory.
Reliably refusing to answer when evidence is insufficient — without losing utility — has only become technically tractable in the last 12-18 months. The architectural primitives didn’t exist before.
82% of healthcare professionals trust clinical AI’s potential — but only 26% actually use it². That 56-point trust-usage gap is the demand pool waiting for a system worth adopting.
Manually edited reference databases can no longer keep pace with the literature. The shift from manual curation to AI-synthesized clinical reference has already begun.
Neither is purpose-built for the oncology decision workflow. The vertical seat is open.
Real-world use by practicing oncologists. Embedded in the visit. Information-only — clear of medical-device regulation.
With user growth — under anonymization and consent — a feedback loop forms: which recommendations were taken, which outcomes followed.
Models tuned on accumulated data deliver patient-specific support — beyond simple search and synthesis.
Once trust is sufficiently earned, formal medical-device approval (FDA / MFDS / CE) — from information to diagnostic AI.
Approval is not the starting point — it is the result of trust sufficiently earned.
Every prescription —
delivered safely.
accurately.
at its best.
Closed beta opening soon. Priority access for medical oncologists across solid-tumor specialties. Additional tumor types added on a rolling basis.
¹ Independent concordance evaluation, Manipal Comprehensive Cancer Centre (peer-reviewed, 2018). ² Wolters Kluwer Health Clinician AI Survey, 2024 — 1,000+ U.S. clinicians. ³ ResearchAndMarkets, AI in Oncology — Global Forecast to 2036 (May 2026). ⁴ DataM Intelligence, AI-CDSS Market (Feb 2026). ⁵ OpenEvidence valuation, 2025. ⁶ Tempus AI public valuation, 2024.