Precision oncology infrastructure
Precision oncology infrastructure for the next generation of clinical software.
UNMIRI builds cross-vendor genomic interpretation, decision support, and literature intelligence APIs for healthtech vendors, biotech medical affairs teams, and the pathologists they serve.
Built on open clinical evidenceHIPAA-ready architectureZero-retention LLM pipelineUS data residency
One platform, three audiences
Built once. Specialized for each buyer.
UNMIRI's shared technical foundation (cross-vendor parsing, a Neo4j knowledge graph, deterministic rendering) supports three distinct surfaces for three distinct audiences.
Cross-vendor NGS interpretation API
Parse Foundation Medicine, Tempus, Caris, Guardant, Natera, and other genomic reports into structured FHIR Genomics output. The Genomics-aware Clinical Decision Support API ships on the same data plane as the natural upsell.
- Vendors
- Foundation, Tempus, Caris, Guardant, Natera, Invitae
- Output
- FHIR Genomics, OncoKB-aligned JSON
- Evidence
- CIViC, ClinVar, ClinicalTrials.gov, openFDA, CPIC
AI-powered literature surveillance
Track every relevant publication, abstract, and trial development across your therapeutic area, with citation-grounded summaries you can defend to your medical leadership.
- Sources
- PubMed, ASCO, ESMO, AACR abstracts, ClinicalTrials.gov updates
- Output
- Citation-grounded summaries with audit-trail provenance
- Scope
- Filtered to your therapeutic area, biomarker set, and competitor pipeline
Free cross-vendor NGS unification tool
Upload reports from any major lab vendor and get unified, citation-backed summaries. Built by software architects with clinical advisor oversight.
- Inputs
- PDFs from any major lab vendor, no integration required
- Output
- Unified summary with variant-level citations and evidence levels
- Pricing
- Free for individual clinical use, no credit card
{
"patient": "synthetic-001",
"tumor": "NSCLC adenocarcinoma",
"variants": [
{
"gene": "EGFR",
"hgvs": "p.Leu858Arg",
"evidence": "OncoKB Level 1",
"drugs": ["Osimertinib"]
}
]
}Sample output
See what cross-vendor unification looks like.
We've built a synthetic NSCLC case showing how UNMIRI's NGS API turns a Foundation Medicine-style report into structured, citation-backed output. The same engine handles Tempus, Caris, Guardant, Natera, and other vendor formats.
Every output field carries provenance: which knowledge-base entry, which RCT, which FDA label. No fabricated drugs, no hallucinated citations.
View the sampleHow UNMIRI works
A three-stage pipeline shared across all four products.
Parsing, reasoning, and rendering are separate layers. Each is replaceable, observable, and testable on its own. The same pipeline powers the API products, the literature platform, and the free pathologist tool.
- 01
Parse
Cross-vendor parsers turn Foundation Medicine, Tempus, Caris, Guardant, Natera, and other vendor reports into a normalized variant model. PDF, structured XML, and HL7 inputs converge on one schema.
AWS Textract handles document extraction. Vendor-specific normalizers map proprietary identifiers to HGVS, transcript-aware coordinates, and OncoKB gene symbols.
- 02
Reason
A Neo4j knowledge graph links variants to evidence: OncoKB levels, CIViC assertions, ClinVar interpretations, ClinicalTrials.gov enrollment criteria, openFDA labels, and CPIC pharmacogenomic guidelines.
Reasoning is graph traversal, not vector similarity. BRAF V600E and V600K stay distinct nodes. Drug-gene-trial relationships are typed edges, not embedding-space neighbors.
- 03
Render
Final output is rendered by deterministic templates against the graph results. LLMs are scoped to extraction edge cases only, never to the clinical recommendation. Every field carries provenance back to its source.
Output formats: FHIR Genomics, JSON, and clinician-facing PDF. Each variant call links to the specific knowledge-base entry, RCT, or FDA label it came from.
Built for clinical accuracy
Architectural decisions you can verify.
All four product surfaces share a common technical foundation designed for the precision oncology workflow. Each decision below is a deliberate move away from patterns that fail in clinical software, with reasoning open to your engineering and compliance teams. We expect to be asked hard questions about every one of them. The answers below should be enough to start the conversation; our security and engineering teams are available for the rest.
GraphRAG, not vector RAG
Cosine similarity conflates clinically distinct variants. BRAF V600E and V600K differ by one amino acid and require different drugs. UNMIRI uses a structured Neo4j knowledge graph instead of a vector store. Variants, drugs, trials, and evidence are typed nodes with typed edges, so a clinician or auditor can replay why each recommendation appeared. Queries like "all FDA-approved drugs targeting EGFR L858R in NSCLC" are graph traversals against typed relationships, not similarity searches over text chunks. The blog post linked below walks through the full failure mode that pushed us off vector RAG and onto a graph in the first place.
Read the technical detailDeterministic, not generative
Final output is rendered by deterministic templates against authoritative sources (CIViC, ClinVar, ClinicalTrials.gov, openFDA, CPIC). LLMs are used only for extraction edge cases, never for the clinical recommendation itself. Identical inputs produce identical outputs. There is no temperature knob on the recommendation, no creative paraphrasing of a drug label, no quiet drift between report runs. Audit logs capture the exact graph state at rendering time, so the same case re-run a year later produces the same recommendation, or a clear delta with reasons.
HIPAA-ready architecture
BAA-eligible cloud services, end-to-end encryption, role-based access controls, audit logging. Three subprocessors total (Vercel, AWS, Anthropic). AWS resources are pinned to us-east-1 by design. PHI is processed in memory and not persisted by UNMIRI after the response is returned. Audit logs capture every variant lookup, drug query, and trial match by request ID, time, and caller. The full subprocessor list, BAA status, and incident-response posture live at /security/subprocessors.
Read about our security postureOpen clinical evidence
Every clinical assertion traces back to a public, citable source: CIViC (CC0 public domain), ClinVar and ClinicalTrials.gov (US Public Health Service data), openFDA drug labels, CPIC pharmacogenomics guidelines, and OncoKB level assignments. Customers can verify any output against the underlying record on the source's own site. Closed-source variant interpretation services keep their reasoning private. UNMIRI does not. License terms and attribution requirements per source are tracked in our internal data-use agreements file and surfaced in every derivative report.
Read more about our approachWho's building this
Built by software engineers, recruiting clinical advisors.
UNMIRI was started by Umair Khan, a software architect with 14+ years of engineering experience, after caregiving for his grandmother through cancer treatment exposed how fragmented the precision oncology software ecosystem is.
The company is in active conversations with multiple board-certified pathologists about formal advisory roles, so engineering decisions can be tested against real clinical workflows before any of it ships to a customer.
Read our story
Building precision oncology software?
We'd love to talk about how UNMIRI's APIs and tools can fit into your stack.
Whether you're a healthtech vendor evaluating the NGS or CDS APIs, a biotech medical affairs team scoping literature surveillance coverage, or a clinician requesting beta access to the free pathologist tool, the form routes you to the right person on the first hop. Pricing, integration scope, and BAA terms are part of the conversation. Reply within one business day.