Why most “precision oncology” integrations take 6 months
Traditional genomic interpretation platforms sell themselves as full-stack replacements: EHR integration, data model migration, custom interfaces, training. The integration timeline reflects that scope — 6–12 months before a single oncologist sees output. Mid-tier labs can't absorb a project like that.
UNMIRI is deliberately a different shape. LIMS-layer integration only. REST API only. Your bioinformatics pipeline, variant-calling, and EHR delivery all stay exactly as they are today. The only new thing in your stack is three API endpoints sitting between your LIMS and your report-delivery workflow.
The three endpoints that make up the integration
| Endpoint | Purpose | Typical call pattern |
|---|---|---|
| POST /v1/reports | Submit NGS report for interpretation | Called once per patient case from LIMS |
| GET /v1/reports/{id} | Retrieve structured JSON insight | Polled or webhook-notified when complete |
| GET /v1/reports/{id}/pdf | Retrieve print-ready 2-page PDF | Called when embedding in report delivery |
Integration patterns we've seen work
Three patterns cover ~90% of our pilot lab integrations. Pick based on your existing report delivery workflow:
- Appendix pattern — UNMIRI PDF is attached as an appendix to your existing NGS report. Lowest risk, fastest to production, preferred during pilot.
- Parallel delivery pattern — Your existing report goes to the oncologist as before; the UNMIRI Actionable Insight is delivered as a separate PDF/email. Lets oncologists opt in to reading.
- Primary-insight pattern — The UNMIRI 2-page Actionable Insight becomes the primary oncologist-facing document; the raw report is archived as the source document. Highest value, typically adopted after pilot validation.
What shadow mode does for you
Default pilot configuration is shadow mode: UNMIRI runs on every NGS case in parallel with your existing interpretation workflow. Your team reviews the UNMIRI output vs. the current output for 30 days. We publish the discordance rate — concordant, partial, discordant — so you have defensible data before any AI-generated content reaches an oncologist. If discordance is unacceptable, you walk away. If it's acceptable, you graduate to one of the three integration patterns above.
HIPAA architecture that makes all of this possible on PHI workflows is covered in Building a HIPAA-Ready Architecture for Clinical Decision Support. The reasoning architecture under the hood is in Why Vector RAG Fails for Oncology.
Volume economics
At $100/report and 200 panels/month, the integration pays for itself within weeks if it prevents one oncology client from migrating to Tempus. The specific numbers for your lab are on the revenue impact calculator. Compliance posture on the security page.
How UNMIRI actually does this
Your LIMS calls three REST endpoints. UNMIRI's pipeline extracts structured variant data (AWS Textract + per-lab parsers), queries a knowledge graph built on OncoKB, ClinVar, ClinicalTrials.gov, and openFDA, and renders a 2-page cheat sheet through deterministic templates. No LLM in the clinical output path. More on the architecture.