Biomedical intelligence infrastructure
BioMedora transforms complex biomedical and clinical documents into structured evidence, longitudinal patient timelines, knowledge graphs, searchable intelligence, and evidence-grounded AI insights — through one privacy-first, API-first platform.
clinical-note.txt · de-identified preview
From document to decision
A single clinical note enters once. It leaves as structured entities, a qualified longitudinal record, a knowledge-graph neighborhood, searchable evidence, and an AI-ready grounding pack.
Structured evidence
Entities with assertion status, temporality, codes, and character-level provenance.
Longitudinal record
Cited chronological timeline — negated and family history correctly excluded.
Knowledge graph
Provenance-carrying relationships: interactions, contradictions, missing follow-ups.
Searchable intelligence
Hybrid BM25 + concept + graph retrieval over section-aware chunks.
Evidence-grounded AI
Answers whose every claim verifies against retrieved source passages.
The full transformation chain — explore it stage by stage below.
Jump to pipelineWhy BioMedora
Generic LLMs answer. NLP libraries extract fields. Search platforms retrieve. Graph platforms store. BioMedora brings the entire chain together — with evidence traceability end to end.
Not just entity extraction. Every fact carries assertion (documented, negated, family history, hypothetical, planned), temporality, coreference resolution, and relationships — because “no evidence of pneumonia” is not pneumonia.
Documents become events; events become a cited chronological timeline per patient. Temporal event splitting separates “diagnosed 2014” from “recurrence 2025” as two distinct clinical moments.
Entities link to biomedical concepts across SNOMED CT, RxNorm, LOINC, ICD, HPO, MedDRA and more, then join a knowledge graph whose every edge carries provenance back to the exact source sentence.
Every AI claim must verify against retrieved evidence with numeric fidelity checks. Unsupported statements are dropped with an uncertainty note — never silently presented as fact.
141 documented endpoints, typed Python SDKs, idempotency keys, cursor pagination, signed webhooks, and streaming chat. One platform, many integrations — from EHR to research pipelines.
De-identification sits inside the pipeline itself: consistent pseudonyms, patient-consistent date shifting, and an independent residual PHI scan that fails closed — blocked documents never proceed on a warning.
Clinical NLP
BioMedora does not depend on an LLM to understand what a clinical document says. A deterministic engine establishes facts; AI reasons over them under contract-enforced guardrails.
medspaCy TargetMatcher & ConText, lexicon fallback. Same input, same facts — every time.
What the document saysSNOMED CT, RxNorm, LOINC, ICD-10-CM, MedDRA, HPO, HGVS, ATC, UCUM — UMLS-license-free by default.
Make it computableLocal open models by default; premium models opt-in. Claims validated against evidence before delivery.
Reason over evidenceDeterministic NLP + terminology linking + governed LLM reasoning = precision, repeatability, traceability, low latency, cost efficiency.
Intelligence pipeline
Explore what happens at every stage of the pipeline — what it does, why it matters, and which API endpoint drives it.
Started metformin 500 mg BID. Creatinine 1.4 mg/dL (high)
{ "text": "metformin", "type": "Medication",
"attributes": {"dose":"500 mg","frequency":"BID"} }Exposed via the API playground — try it live.
Patient timeline
Events compile into a cited chronological record grouped by year — filterable, inspectable, and accessible. Select any event to see exactly which sentence it came from.
2019
2020
2021
2022
2023
2024
Knowledge graph
Every node and edge traces back to its source sentence. Click nodes to inspect concepts, dates, confidence, citations — including detected interaction risks surfaced from the knowledge base.
Evidence-backed AI
Every generated claim is verified against retrieved evidence with numeric fidelity checks. Unsupported statements are dropped with uncertainty notes — and claim verdicts are explicit.
Ask a question
Simulated interface mirroring the production contract of POST /v1/reasoning/query. Run it against your own tenant in the playground.
Two independent signals support a decline in renal function between 2020 and 2021, with an important caveat about follow-up:
Creatinine rose from within normal limits (1.1 mg/dL, 2020-02) to 1.6 mg/dL flagged HIGH on 2021-09-22.
The 2021 report explicitly recommended a repeat renal panel in 90 days.
Metformin dosing was not re-evaluated in retrieved documents after the creatinine increase.
Absence of evidence in the current repository — not evidence of absence.
Uncertainty: No documents dated after 2021-12 were found referencing repeat renal testing — flagged as missing follow-up by data-quality analysis.
Evidence
[1] DOC-19011 · p.3 · Labs
“Basic metabolic panel: creatinine 1.1 mg/dL (normal).”
[2] DOC-20455 · p.3 · Labs
“Creatinine 1.6 mg/dL — HIGH. Repeat renal panel in 90 days.”
Product tour
Explore a current preview of the BioMedora workspace, or open the running frontend for the complete functional experience.
Opens the BioMedora application in a new tab.
e2e-admin
Admin
Your biomedical intelligence workspace
Document pages processed
18,204
6,796 of 25,000 remaining
AI reasoning requests
847
4,153 credits remaining
Document repository
62 GB
100 GB included
Total logical data
97 GB
Documents + structured + knowledge
Data platform
View detailsAI intelligence
Every AI answer is grounded in retrieved evidence with citations.
Solutions
The same pipeline — privacy, understanding, structure, evidence — powers very different missions. Each workflow maps to real platform capabilities.
If your organization moves biomedical documents, BioMedora can compile them. Tell us about your data and we will map the pipeline to it.
Talk to BioMedoraAPI-first
Your EHR, LIMS, research platform, data lake, or AI assistant consumes BioMedora through one governed API surface — 141 documented endpoints, typed errors, streaming chat, and signed webhooks.
Create a tenant account and issue a bearer token or scoped API key — secrets shown once, stored hashed.
pip install biomedora-sdk — typed resources for patients, documents, deidentification, reasoning, billing and more.
POST /v1/documents with an Idempotency-Key; processing stages are observable via the jobs API.
Entities, assertions, temporal events, timelines, graph neighborhoods — every item traceable to its source sentence.
pip install biomedora-sdk
from biomedora_sdk import Client
client = Client(api_key="YOUR_API_KEY",
base_url="https://api.biomedora.com")
# 1. Compile a document into structured intelligence
doc = client.documents.create(
content=open("discharge_summary.txt").read(),
patient_id="patient_123",
)
# 2. Read the cited longitudinal timeline
timeline = client.patients.timeline(patient_id="patient_123")
# 3. Ask an evidence-grounded question
answer = client.reasoning.query(
patient_id="patient_123",
question="What changed after the medication was started?",
)
for claim in answer["claims"]:
print(claim["status"], claim["confidence"], claim["text"])Also distributed as biomedora-client — identical API.
curl -X POST https://api.biomedora.com/v1/nlp/analyze \
-H "Authorization: Bearer $BIOMEDORA_TOKEN" \
-H "Content-Type: application/json" \
-H "X-Request-ID: req-0042" \
-d '{
"content": "Started metformin 500 mg BID. No evidence of pneumonia."
}'// TypeScript — REST integration (typed fetch)
const res = await fetch("https://api.biomedora.com/v1/reasoning/query", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.BIOMEDORA_TOKEN}`,
"Content-Type": "application/json",
"Idempotency-Key": crypto.randomUUID(),
},
body: JSON.stringify({
patient_id: "patient_123",
question: "What evidence suggests renal deterioration?",
}),
});
const answer = await res.json(); // claims[], citations[], uncertaintySecurity & privacy
BioMedora is designed for privacy-sensitive biomedical workflows: privacy controls are embedded throughout the data lifecycle, from ingestion through AI dispatch.
Full security architectureDetection → pseudonymization → patient-consistent date shifting → independent residual scan. Any high-confidence PHI residue blocks the document outright — empty output, never a warning.
Server-derived tenant identity on every request; patients namespaced per tenant; cross-tenant access denied and audited — including for administrators.
Deterministic NLP first, local models by default. External LLMs require minimized PHI and explicit operator opt-in — and are disabled entirely in BYOC/on-premises deployments.
Append-only, hash-chained audit records for every sensitive action — who, what, when, outcome, request ID — with verification that pinpoints any tampered entry.
Single-tenant cloud by default; runs in your AWS account (BYOC) or fully on-premises with in-hospital inference and no external egress.
Document content is untrusted evidence. Instructions hidden inside clinical text are surfaced transparently and never forwarded to models as directives.
FAQ
A biomedical intelligence platform that compiles heterogeneous clinical and research documents into structured evidence, longitudinal patient timelines, knowledge graphs, and searchable intelligence — then runs evidence-grounded AI reasoning on top, all through one API-first surface.
Tell us about your document landscape and we will map BioMedora to your workflows — a tailored walkthrough, not a sales script. Or start immediately in the API playground.
No fabricated case studies — judge the product itselfSynthetic demo data onlyYour inquiry details stay with us
Prefer to explore first? Open the playground or read the docs.