Loading...
Please wait while we fetch your data
Turn clinical notes
into knowledge.
Up to 80% of clinical data lives in unstructured text. Sulci extracts, normalizes, and connects it into knowledge your AI can reason over — in minutes, not months.
60x Faster Abstraction*96.7% Clinical QA Accuracy*Assertion-Aware "denies" ≠ "has"85% Signal Captured*<50ms P95 Latency*
*Based on internal benchmarks. Results may vary by document type and complexity.
app.sulci.ai
clinical_note_2847.txtProcessing
Assessment: 62 y/o male presents with acute exacerbation of CHF.
History of T2DM on metformin 1000mg BID, HTN controlled with lisinopril 20mg daily.
Patient denies chest pain. EF 35% on last echo. BNP elevated at 1,847 pg/mL.
Plan: IV furosemide 40mg, cardiology consult, repeat echo in AM.
extracted_concepts8 found
ConditionCHFSNOMED:42343007
ConditionT2DMOMOP:201826
DrugmetforminRxNorm:6809
ConditionHTNSNOMED:38341003
DruglisinoprilRxNorm:29046
ObservationEF 35%LOINC:10230-1
LabBNP 1847LOINC:42637-9
DrugfurosemideRxNorm:4603
Team with experience at
HCA HealthcareCommureUCF HealthAugmedix + Google
The problem
Your smartest data is your dumbest asset.
LLMs can generate text. They can't reason over your clinical data without a knowledge backbone.
$13B+
estimated annual cost of manual chart abstraction
Industry estimates suggest health systems spend billions on human abstractors reviewing notes for quality measures, risk adjustment, and research.
72h
typical chart abstraction time per study
Manual review per patient for quality measures, research, and coding can take days. At scale, this means months of delay per study.
~30%
of clinical signal estimated to be missed
Negation, temporality, and context are frequently lost in manual abstraction, leading to significant downstream rework and cost.
Core Engines
Three engines. One pipeline.
From raw clinical text to a queryable knowledge graph in three steps.
Extract
Clinical NLP
Rule-based and transformer ensemble pipelines that extract clinical mentions from unstructured notes. Assertion, negation, temporality, and experiencer built in.
Named entity recognition for conditions, meds, procedures
Assertion classification (present, absent, possible)
Section-aware processing (HPI, Assessment, Plan)
Temporal relationship extraction
Normalize
Ontology Mapping
Map extracted concepts to UMLS, OMOP, and custom ontologies with candidate ranking, confidence scoring, and full provenance tracking across the entire UMLS Metathesaurus.
Full UMLS Metathesaurus (SNOMED, ICD-10, LOINC, RxNorm, MeSH...)
OMOP CDM concept mapping with vocabulary versioning
Ranked candidates with confidence scores
Custom ontology support for institution-specific codes
Connect
Knowledge Graph
Build queryable patient knowledge graphs from normalized facts. Temporal relationships, drug-condition links, and lab trajectories — all connected and traversable.
Patient-centric graph with clinical edges
Drug-condition and drug-drug interactions
GraphRAG natural language querying
Neo4j persistence + in-memory projection
How it works
See the transformation
From a messy clinical note to structured, ontology-mapped clinical facts — in one API call.
InputRaw Clinical Note
Pt is a 62yo M with hx of T2DM, HTN, and CKD stage 3. Currently on metformin 1000mg BID and lisinopril 20mg daily. A1c 8.2% on last check. No chest pain.
OutputNormalized Clinical Facts
CONDType 2 diabetes mellitusC0011860
CONDEssential hypertensionC0085580
CONDChronic kidney disease, stg 3C2316810
DRUGmetformin 1000 MG OralC0987664
DRUGlisinopril 20 MG OralC0689142
MEASHbA1c 8.2%C0474748
CONDChest painnegC0008031
Developer Experience
Ship in minutes,
not months.
First-class SDKs for Python and TypeScript. Extract clinical concepts, map to UMLS and OMOP ontologies, and query your knowledge graph with a single API call.
One-line pipeline execution
Built-in PHI de-identification
Real-time streaming support
High-availability architecture
200 OK · 47ms
{
"facts": [
{ "cui": "C0011860", "name": "T2DM", "confidence": 0.97 }
{ "cui": "C0085580", "name": "HTN", "confidence": 0.95 }
],
"graph_nodes": 8, "graph_edges": 10
}
v1.4.2
1import sulci23client = sulci.Client(api_key="sk_live_...")45# Run the full extraction pipeline6result = client.pipeline.run(7 input="Pt is a 62yo M with hx of T2DM, HTN.",8 ontologies=["umls", "omop", "snomed"],9 include_graph=True10)1112for fact in result.facts:13 print(f"{fact.cui}: {fact.concept_name}")
Platform
Built for the full clinical stack
From ingestion to interoperability — every layer of clinical data processing in one platform.
FHIR R4 Interoperability
Import and export complete patient records as FHIR R4 Bundles. Conditions, medications, observations, and procedures mapped automatically.
ConditionsMedicationsObservationsProceduresAllergiesImmunizations
Clinical Trials
Automated eligibility screening. Cohort matching against I/E criteria with patient ranking.
Drug Safety
Adverse event detection, pharmacovigilance signals, and drug interaction checking.
Billing & Coding
ICD-10, CPT, DRG code suggestion from clinical facts. Automated coding validation.
Medications & Dosing
RxNorm mapping with interaction checking and clinical calculator integration.
What teams say
Trusted by clinical data teams
“The pre-charting alone saves me 20 minutes per patient. Having structured clinical decision support with full provenance — knowing exactly where every recommendation came from — changes how I practice.”
DCH
Dr. Cindy Hird
Emergency Medicine Physician
“I was blown away by what this can do. Extracting concepts from messy notes, mapping them to OMOP, and building a queryable graph — all in one pipeline. This is the clinical data tool I've been waiting for.”
DJS
Dr. Jose Santini
Nephrologist
“Provenance and reasoning are the holy grail of responsible AI in healthcare. Knowledge graphs are key to that, and Sulci nails the connection between unstructured notes and structured, traceable clinical facts.”
DWH
Dr. Waqar Haider
Emergency Medicine Physician
Testimonials reflect individual opinions and experiences. Results may vary.
Standards
Full UMLS Metathesaurus and beyond.
Native support for the entire UMLS Metathesaurus, OMOP CDM, and custom ontologies. Over 200 source vocabularies, zero custom adapters.
UMLS MetathesaurusOMOP CDM v5.4FHIR R4SNOMED CTICD-10-CMLOINCRxNormMeSHCPTNDCCDS HooksSMART on FHIR
Security & Compliance
Designed for regulated environments.
Security-first architecture with audit logging, encryption at rest and in transit, and role-based access controls.
HIPAA Ready
Designed for compliance
SOC 2 Type II
In process
BAA
Available on request
FHIR R4
Standards-based interop
OMOP CDM v5.4
Standard data model
21 CFR Part 11
Architecture aligned
Designed to integrate with your infrastructure
Epic
Cerner
Carequality
CommonWell
Veeva Vault
Databricks
Redox
Health Gorilla
MEDITECH
OpenEHR
Founder
Built by a physician who ships.
Alex Stinard, M.D.
Founder & CEO
“Unstructured notes are a smooth brain. I've spent 20 years watching clinical signal get lost in free text. Sulci adds the folds.”
Emergency physician executive and clinical AI architect. Previously at Commure, designed and governed agentic ambient scribing at national scale — multi-agent orchestration with knowledge-graph memory, ontology-aware prompting, and deterministic EHR tool use.
Enterprise SME for clinical documentation at HCA Healthcare. Led the Meditech Expanse rollout affecting millions of ED visits. Vice Chair of the Board at Lake Monroe Hospital. Assistant Professor of Emergency Medicine at UCF.
CommureClinical AI Architect
HCA HealthcareEnterprise SME, Clinical Documentation
Lake Monroe HospitalVice Chair of the Board
UCFAsst. Professor, Emergency Medicine
Multi-Agent SystemsKnowledge GraphsOntology DesignFHIR/HL7Safety & Red-TeamingCDI/DRG
Pricing
Start free, scale with us
Explore the platform at no cost. Request early access when you're ready to build.