Do you integrate with Surescripts?+
Yes. We're a Surescripts authorized integration partner. Your clinical GraphRAG system connects directly to the Surescripts network for e prescribing, prior authorization, benefits verification, and medication history: so your AI doesn't just answer clinical questions, it takes action on the workflows that used to take weeks.
Is your knowledge graph HIPAA compliant?+
Yes. We sign BAAs, implement property level access controls, encrypt in transit and at rest, log every query, and deploy in your VPC or on prem when your Security Officer requires it. The platform is SOC 2 Type II audited.
How does GraphRAG reduce clinical hallucinations?+
GraphRAG grounds every answer in your clinical knowledge graph. The model can only cite entities that exist in the graph, and every fact carries provenance back to the source system it came from. Clinical evaluations we've run typically show single digit percentage hallucination rates versus 10 to 20% for vector only baselines: and we guarantee measurable hallucination reduction in writing.
How does this work with FHIR and SNOMED CT?+
FHIR resources map naturally to graph nodes and edges: Patient, Condition, MedicationRequest and their references are essentially graph structure already. We ingest FHIR as is and layer SNOMED CT (plus ICD 10, RxNorm, LOINC) as semantic tags on the nodes, so queries can walk both instance data and terminology hierarchies.
Can we run this without moving PHI to a third party?+
Yes. Every core technology in the stack (Neo4j, Stardog, TigerGraph, Apache AGE) supports on premises deployment. For GraphRAG we support open weight models (Llama, Mistral) hosted in your VPC or on prem GPUs, with zero retention API agreements when public LLM APIs are on the table. Your data stays within your infrastructure.
What's a realistic first engagement for healthcare AI?+
A 6 week pilot with one clinical use case: GraphRAG powered clinical Q&A or prior auth automation are common starts. Fixed scope, outcomes guaranteed in writing: hit the agreed hallucination reduction target or the pilot is free.
What's the difference between HL7 v2 and FHIR: and do you support both?+
HL7 v2 is the older pipe delimited messaging standard still running most hospital interfaces (ADT, ORM, ORU); FHIR (Fast Healthcare Interoperability Resources) is the modern REST/JSON API standard built on discrete resources like Patient, Condition, and MedicationRequest. We support both: we ingest HL7 v2 through interface engines like Mirth Connect and consume FHIR R4/R5 via SMART on FHIR and CDS Hooks: then map both into one knowledge graph so an LLM can reason across legacy and modern data with full provenance.
Can you integrate our clinical AI with Epic, Cerner (Oracle Health), or Athenahealth?+
Yes. We connect to Epic, Cerner/Oracle Health, Athenahealth, eClinicalWorks, NextGen, and Veradigm through their FHIR APIs, SMART on FHIR launch, and HL7 v2 interfaces: plus C CDA exchange and HIE networks (Carequality, CommonWell, TEFCA/QHIN) where a direct EHR API isn't available. Your knowledge graph becomes the vendor neutral layer, so a GPT- or Llama based assistant can read Epic and Cerner data through one grounded interface instead of one brittle integration per system.
What is the NCPDP SCRIPT standard, and do you support e prescribing?+
NCPDP SCRIPT is the national standard for electronic prescription messaging (new prescriptions, renewals, cancellations, prior authorization, and medication history) and it's the language the Surescripts network speaks. As a Surescripts authorized integration partner we transact on NCPDP SCRIPT for e prescribing, electronic prior authorization (ePA), real time prescription benefit (RTPB), formulary and benefits checks, and medication history, with EPCS ready controlled substance and PDMP workflows.
How does RxNorm map to NDC, and do you normalize drug data?+
RxNorm is the normalized clinical drug vocabulary (ingredient, strength, dose form); NDC (National Drug Code) identifies the specific manufactured, packaged product. One RxNorm concept typically maps to many NDCs, so drug logic lives at the RxNorm level while dispensing and billing happen at the NDC level. We model that RxNorm ↔ NDC hierarchy natively in the graph and enrich it with SNOMED CT, LOINC, and ICD 10: so interaction checks, formulary logic, and dispensing all resolve to the same canonical drug entity.
Do you use openFDA, DailyMed, and FAERS for drug and safety data?+
Yes. We enrich the drug graph with public FDA sources (openFDA APIs, DailyMed structured product labeling (SPL), the NDC Directory, and FAERS adverse event reports) and can layer in a client's licensed commercial databases (First Databank, Medi Span, Micromedex) where available. That gives contraindication, interaction, and pharmacovigilance signals a single provenance tracked source instead of scattered lookups.
Is your clinical AI a HIPAA compliant LLM, and how do you handle PHI de identification?+
Yes. We run GraphRAG on open weight models (Llama, Mistral) inside your VPC or on prem GPUs so PHI never leaves your infrastructure, with BAAs, property level access controls, encryption in transit and at rest, and full query audit logging: SOC 2 Type II audited. When a public LLM API is in scope we use zero retention agreements and a PHI de identification layer (Safe Harbor / Expert Determination) so only de identified context is ever sent off premises.