Questions, answered.
The cross-cutting questions buyers, partners, regulators, and families ask first. Architecture, pilot, pricing, validation, security, patient, investor, and how Burna compares to the systems already in your stack. Audience-specific FAQs live on the audience pages.
Product · Architecture · Pilot · Pricing · Validation · Security · Patient · Investor · Differentiation. Jump anchors below.
WHO-UMC and Kramer attribution. Multi-drug oncology, not single-drug lookup.
CTCAE v5.0 and v6.0. The full ontology, structured for grading.
Filed. Twelve specialized agents in a cascading constraint pipeline. Citation-bound by design.
What Burna does.
What is Burna AI, in one sentence?
Burna is the safety and data quality platform for oncology drug development, from clinical trials through postmarket surveillance. It produces CTCAE coding, including TERM and CATEGORY, coupled with intelligence to support determination of severity GRADE, with mandatory citation to source text and multi-drug attribution using WHO-UMC and Kramer algorithms across 42 oncology regimen profiles. A clinician reviews and signs every record.
What does the engine actually do?
The engine reads the clinical note, the labs, the medication list, the prior cycles, and the regimen protocol. It surfaces adverse events with the source sentence highlighted, the CTCAE criterion quoted, and the WHO-UMC or Kramer attribution path attached. The clinician reviews, modifies if needed, and signs. The signed record flows into the EDC or the postmarket ICSR workflow. Every step is audit-ready.
What are the product lines?
Four. CTCAE Grading for cancer centers and academic medical centers. Protocol Safe for pharma sponsors who need an isolated AI environment inside their own cloud. Postmarket Pharmacovigilance for sponsors managing safety data after approval. Care Journal for patient self-reporting between visits. The four products run on the same engine and share the same citation discipline.
What is Protocol Safe?
Protocol Safe is a sponsor-tenant deployment of the Burna engine inside the sponsor’s own AWS, Azure, or GCP environment. The trial protocol does not leave sponsor infrastructure. A retrieval-augmented agent inside the sponsor’s cloud receives queries from the Burna engine and returns only structured attribution signals (probability scores, citation chains), not the protocol content. Content Boundary Middleware enforces fail-closed validation. Protocol text, excerpts, and any string over 50 characters are blocked at the boundary. Full architecture in the 8-page Protocol Safe technical brief. See /pharma for the sponsor view.
What is the Care Journal?
The Care Journal is the patient-facing surface. Patients on an oncology trial can share what is happening between visits, in their own words and in their own language, through text, voice, or short video (CareVid). Those reports flow to the care team as structured PRO-CTCAE input, not as raw inbox noise. The patient view is at /patients.
What is Postmarket PV?
Postmarket Pharmacovigilance is the same engine applied to ICSR intake, deduplication, MedDRA coding, first-pass grading, and narrative drafting for drugs already on market. The 21 CFR 314.80 obligation is met by construction; the audit trail is complete by definition. Launches 2027. Pilot partners scoping now lock in 2026 economics.
Is Burna a medical device?
No. Burna is a clinical decision support tool. Clinicians approve every output. The platform does not prescribe, modify dose, or alter care pathways. 21 CFR Part 11, SOC 2, HIPAA, and GDPR aligned.
The cascading constraint pipeline.
What is the engine?
The proprietary Burna AI engine is a patented cascading constraint pipeline of twelve specialized agents. Each agent operates under the constraints of every upstream agent. The output space narrows at every step. The system cannot produce a grade outside the defined CTCAE set, cannot skip citation, and cannot contradict its own upstream findings. Two patents filed. The full agent-by-agent breakdown is at /engine.
What does "citation-bound" actually mean?
Every suggested grade requires a source sentence in the clinical note and a matching CTCAE criterion in the retrieval index. Without both, the system returns "no grade available." Architectural constraint, not runtime filter. The model is structurally prevented from producing grades it cannot cite.
How does Burna handle hallucination?
Constraint architecture, not statistical guarantee. The output space is restricted to valid CTCAE grades. The engine cannot return a grade without a source citation. Constraint enforcement is structural, not a transparency layer added after generation.
What happens when the model is unsure?
Fail-closed. Low-confidence outputs are surfaced to the clinician with the constraint failure flagged. The system does not produce an ambiguous grade.
Why TERM and CATEGORY before GRADE?
Because the same numerical finding can mean two different things depending on the CTCAE category. An ejection fraction decline under "Investigations" is a measurement artifact; under "Cardiac Disorders" it is drug-induced heart damage. Without category-aware coding, a safety pipeline either over-reports cardiac events or under-reports them. Burna performs TERM and CATEGORY identification before grade determination, so the same numerical finding routes to the correct organ system class with full citation to source text.
What are the two patents on?
The cascading constraint pipeline architecture and the citation-bound retrieval method that enforces it. Filed in 2025. The patent posture is part of the architectural moat. Full architecture at /engine.
Why twelve agents and not one large model?
A single model does classification. The engine does reasoning under constraint. Twelve agents allow each step to bound the next, so the output narrows toward a defensible grade rather than expanding into a plausible one. The pipeline spans term identification, category assignment, severity grade determination, multi-drug attribution, citation, and confidence calibration. Each step is constrained by every step before it.
How does real-time work across multiple sites?
The engine processes grades inside the SMART on FHIR app at each site as the encounter happens, and the cross-site consistency surface aggregates in real time. Cross-site grading inconsistency and attribution drift surface while the trial is running, not at the next monitoring visit. This is what makes the FDA-NCI "sub-optimal, unreliable, inefficient" characterization addressable at the architectural layer.
The 90-day design partner pilot.
What does a Burna design partner pilot look like?
A 90-day, four-phase engagement. Phase 1 (weeks 1 to 3) is scoping: workflow study co-design, residency configuration, IRB review, MSA and BAA execution. Phase 2 (weeks 4 to 6) is integration: SMART on FHIR registration, OAuth provisioning, EDC integration with Medidata Rave, Veeva Vault EDC, or Oracle Clinical One, Implementation Lead on-site. Phase 3 (weeks 7 to 18) is in-pilot: production grading against pre-registered endpoints, paired-grader study, weekly check-ins. Phase 4 (weeks 19 to 21) is readout: database lock, statistical analysis, manuscript co-drafting, decision on enterprise contract.
What are the pilot outcome terms?
In success, the pilot delivers citation-bound CTCAE grading inside the EHR, the labor recovery the platform was scoped to produce, and a peer-reviewed manuscript with the institution as primary authors. If pre-registered endpoints are not met at study end, the contract terminates with no further fee, and the institution retains the deployment, the training materials, the change-management playbook, the Implementation Lead’s work product, and the co-authored manuscript. Either way, the institution is the named author on a peer-reviewed paper.
Who co-designs the endpoints?
Pilot endpoints are co-designed with Dr. Andrea Pirzkall (formerly Chief Medical Officer at Replimune and Asher Biotherapeutics; Executive Director Clinical Development at BeiGene; Principal Medical Director at Genentech) before kickoff. Paired-grader timing, documentation completeness, and multi-drug attribution agreement are typical endpoints. Specifics are co-designed with the unit; the science is the procurement asset.
What ships with every pilot?
Six artifacts. The citation-bound CTCAE grading platform itself. The 12-pillar Implementation Playbook (LMS-compatible training modules, communication templates for medical executive committee and IRB, future-state workflow maps, governance and policy guidance). An on-the-ground Implementation Lead named in the contract. Pre-registered validation co-authorship. The five-page Architecture and Implementation Brief, BAA, and residency configuration. Pilot outcome terms. Detailed at /cancer-centers §5.
How long does integration take?
Two to three weeks from kickoff to first signed grade in production. The customer’s IT provisions one OAuth client and one FHIR endpoint. Burna’s engineering team handles FHIR mapping, residency configuration, and EDC integration.
When does the 2026 design partner cohort close?
The 2026 design partner cohort accepts pilot scoping conversations through the end of Q3 2026. Pilots that begin in Q4 2026 ship co-authored manuscripts in late 2027.
What does a scoping conversation look like?
Fifteen minutes for first contact. Sixty to ninety minutes for the technical and clinical scoping session if there is fit. Workflow study endpoints are co-designed in Phase 1 of the pilot, before any contract is signed.
How Burna prices.
How does Burna price?
Three motions. Per-seat for cancer centers and academic medical centers. Per-trial for pharma sponsors. Per-trial or white-label for CROs. Postmarket pharmacovigilance prices per ICSR with an enterprise minimum.
What does it cost for a cancer center?
Mid-size academic medical centers: $200K to $500K annually, recommended for 30 to 80 active oncology trials and 4 to 12 research coordinators. Major comprehensive cancer centers: $500K to $1.5M annually, recommended for 100+ active oncology trials, multiple sites, and dedicated oncology pharmacovigilance teams. Pilots are zero cost during the 2026 design partner cohort. Regional pricing available for centers in emerging markets where US pricing is unworkable. Multi-year contracts include built-in price protection through 2028. See /cancer-centers §7 for the full breakdown.
What does it cost for a pharma sponsor?
Protocol Safe per-trial: $100K to $250K. Portfolio pricing for sponsors running three or more active oncology programs: $2M to $5M ACV. Postmarket PV at $15 to $50 per ICSR at blockbuster-class volumes, $500K to $2M enterprise minimum. Pilot partners scoping now lock in 2026 economics. See /pharma §10 for the full breakdown.
What does it cost for a CRO?
Per-trial standard: 90-day paid pilot at $25K to $50K, with the first 30 days at no charge to remove procurement friction. Conversion to a 12-month annual contract at $75K to $150K per trial once pilot success metrics are met. Top-6 CRO white-label engagements scoped separately from $400K to $20M annually. See /cros §12 and §13 for the full breakdown.
How much can Burna reduce query labor?
Per-trial query labor at industry-standard oncology rates is $192K to $710K (Oracle ClearTrial, 2021). Approximately 85.7% of those queries are confirmation requests, not data corrections (Pronker et al., British Journal of Clinical Pharmacology, 2011). Burna’s citation-bound architecture closes those confirmation queries structurally by attaching the source sentence, the CTCAE criterion, and the WHO-UMC or Kramer attribution path to every grade. The query the monitor would have opened is answered before it is opened. The 90-day pilot’s pre-registered endpoints typically include a target of 50% reduction in confirmation queries on adverse event forms; the exact figure depends on the current baseline at your unit, which the scoping conversation surfaces.
Is the Explorer Tier a CTA?
The Explorer Tier is a free product offering for individual oncologists, research coordinators, and clinical fellows. 50 grading runs per month, CTCAE v5.0 coverage, single-drug attribution, watermarked exports, no PHI, no EHR integration. It is a product offering, not a site-wide CTA. The site-wide CTA is the 30-minute conversation. The transition path from Explorer to the full platform is documented when an institution is ready.
How Burna proves it.
What validation evidence is available today?
Ongoing internal testing demonstrating strong agreement with expert clinicians. Three pre-registered studies in flight: a prospective workflow study at a US cancer center, a 1,200-chart retrospective accuracy study at a US NCI-designated comprehensive cancer center, and a second-site prospective workflow pilot inside an active GI oncology and Phase 1 trials program. A postmarket pharmacovigilance pilot launches in 2027. Full study-by-study detail at /validation.
Why are specific accuracy percentages not on the public site?
Conservative reporting until peer review supports the claim. Specific percentages and kappa numbers are available under NDA to pilot scoping conversations, investors in diligence, and regulatory advisors. The first peer-reviewed publication is the public moment for those numbers.
What are the target venues and the publication timeline?
Bio-IT World 2026 poster (April 2026). JCO Clinical Cancer Informatics manuscript target 2026 to 2027. Annals of Oncology manuscript target 2027. Regulatory science journal (postmarket PV pilot) target 2027 to 2028. Pre-prints on medRxiv within 60 days of database lock.
What is the approach to negative results?
Published. Every time. A cohort, a regimen, a language, or a workflow that does not perform as hypothesized is reported. Failed studies still publish. Non-negotiable.
How is validation different from a vendor benchmark?
Pre-registered endpoints. Peer-reviewed target venues. Independent IRB or HREC oversight. External co-authorship from the partner site. Pre-print on medRxiv within 60 days of database lock. The validation study endpoints are the same endpoints a sponsor’s CMO would want to see before pilot kickoff. The science is the procurement asset, not a marketing artifact.
Will Burna train on customer data?
Not without explicit opt-in and governance review. Each model update ships with a model card, eval results against a standard benchmark, and a rollback path to the previous model version. Updates are opt-in, not automatic.
What about data poisoning during validation?
Every validation dataset is held out from training. Model freeze at study start. Validation-dataset text is never sent to a model that has seen or will see that text during training. Architectural, not procedural.
HIPAA, 21 CFR Part 11, GDPR aligned.
What compliance frameworks does Burna align to?
HIPAA (BAA executed per engagement). 21 CFR Part 11 (validation package ships with every engagement, e-signature trail built in). GDPR, UK GDPR, PIPEDA, APPI, PIPA, Australian Privacy Act, LGPD. GVP Module VI for pharma engagements. SOC 2 Type II audit kicked off Q2 2026; Type I attestation available now. ISO 27001 mapped, formal certification target 2027. Full matrix at /it-and-security.
What is the FDA posture?
Burna AI is eligible for the CDER Emerging Drug Safety Technology Program (EDSTP). The EDSTP application is in review. The program was established for exactly this category of upstream safety infrastructure. The 2019 FDA-NCI workshop characterized the current state of oncology adverse event attribution as sub-optimal, unreliable, inefficient; the agency is engaged on solutions that match the architectural profile Burna ships.
What about the EU AI Act?
Monitored. High-risk AI system provisions reviewed quarterly with EU regulatory counsel. Clinical Decision Support Tool positioning and human-in-the-loop architecture are designed for compliance.
Where does patient data go?
Patient identifiers stay inside the customer’s network in on-premises deployments and inside the customer’s jurisdiction in regional cloud deployments. The grading model only sees de-identified clinical text. Architecture diagram and data flow documentation available in the five-page Architecture and Implementation Brief.
Which regions are supported?
US (AWS, Azure), EU (Frankfurt, Dublin, Paris), UK (London), Canada (Toronto), Australia (Sydney), Japan (Tokyo), Korea (Seoul), Singapore, UAE (Dubai), KSA (Riyadh) at launch. Sovereign cloud partnerships in negotiation for Oracle Australia, G42 UAE, and STC Saudi Arabia. Data does not cross jurisdictions without explicit configuration.
Which EHRs and EDCs are supported?
SMART on FHIR inside Epic, Oracle Health (Cerner), Athena, Allscripts, eClinicalWorks, MEDITECH, Dedalus Orbis, AGFA, and regional APAC systems. If the EHR supports FHIR R4, Burna runs inside it. EDC integration with Medidata Rave, Veeva Vault EDC, Oracle Clinical One, and Castor at launch.
What contracts and legal artifacts are pre-prepared?
Master Service Agreement, Business Associate Agreement, Data Processing Agreement, subprocessor list, insurance certificates, right-to-audit clause, source code escrow option, and 90-day exit and data export provisions. Single-document downloads from the procurement portal. General Counsel review measured in weeks, not quarters.
Is Burna covered by insurance?
Yes. Cyber liability and professional liability coverage in line with healthcare AI standards. Certificates available on request.
How does a CISO review actually start?
Email security@burna.ai for the five-page Architecture and Implementation Brief. A mutual NDA is available before any data flow conversation. The standard CISO timeline runs 4 to 16 weeks; Burna’s deployment phasing accommodates extended review without breaking the integration architecture. The CISO view is at /it-and-security.
The Patient Portal.
What is the Patient Portal?
A way for patients on an oncology trial to share what is happening between visits, in their own words and in their own language, through text, voice, or short video. The care team receives structured PRO-CTCAE input rather than inbox noise, with the patient’s own words preserved in the record.
Is using the Patient Portal required?
No. Reporting through the portal is usually optional. The trial team may use it to make their work faster; the coordinator can explain how it fits into a specific trial.
Who sees what a patient reports?
The patient’s trial care team (coordinator, trial nurse, physician) and no one else by default. The hospital’s standard patient privacy rules apply. HIPAA in the US, GDPR in Europe and the UK, equivalent in every jurisdiction Burna operates in.
What about a medical emergency?
Call the trial’s 24/7 line, the doctor’s office, or the local emergency number. Burna is not an emergency tool. It is a tool for communication between visits.
What languages are supported?
The patient portal works in seven languages at launch, with more in pilot. If a language is not yet supported, the coordinator can request it; languages are added by request, with translators who specialize in oncology terminology. Clinical notes are processed in nine languages at launch (English, French, German, Spanish, Portuguese, Italian, Japanese, Korean, Arabic).
How is consent handled?
Consent for the Patient Portal is part of the trial’s standard informed consent process. The portal does not change the trial’s data governance; it adds a structured channel inside it. The patient view is at /patients.
The seed, the Series A, the exit.
What is the current round?
Physician-led seed. $3M SAFE at $30M cap, 20% discount, MFN. Mechanism is WeFunder Reg D 506B (accredited and non-accredited physician LPs welcome). Minimum check $25K. Self-service for checks $50K and below; no NDA, no data room, no negotiation on terms. The current SAFE round closes June 30, 2026.
When is the Series A?
Q1 2027 target at $200M to $300M post-publication. Three pre-registered studies. One peer-reviewed publication. The Series A is the readout.
What is the investment thesis?
The validation roadmap is the investment thesis. Citation-bound architecture with named-algorithm attribution operates upstream of the cases that downstream PV systems manage. The architectural moat compounds while validation publishes. Comparables in healthcare AI infrastructure (OpenEvidence at $12B Series D, Hippocratic AI at $3.5B Series C, Abridge at $5.3B Series E, Suki AI at $500M Series D, Tempus AI at ~$8.1B market cap) suggest the white-space play in oncology safety has room. Burna is positioned as AI-native and two-sided where Flatiron was pre-AI and single-sided. The investor view is at /investors.
What is the regulatory risk?
Burna is a clinical decision support tool. Clinicians approve every output. The platform does not prescribe, modify dose, or alter care pathways, so it does not trigger FDA Software as a Medical Device (SaMD) regulation. 21 CFR Part 11 e-signature trail built in. HIPAA and GDPR aligned. SOC 2 Type II audit kicked off Q2 2026. CDER EDSTP application in review. Dose management AI and RECIST were considered and rejected at the platform-shape level precisely to keep the regulatory surface narrow.
What is the exit path?
Three credible paths. Strategic acquisition by a pharma-services or clinical-trial-infrastructure incumbent (Veeva, Medidata, ICON, IQVIA, Parexel, Labcorp Clinical Trials). Strategic acquisition by a pharma sponsor or platform building safety-data capability in-house (the Flatiron-to-Roche pattern, adapted for AI-native and two-sided). Public market path via comparables. The Flatiron precedent at $1.9B in 2018 sets a conservative floor; the AI-native and two-sided model expands the addressable surface.
How are advisors structured?
25 advisors across 5 councils. Council leads include former VP at Celgene, former CMO at Replimune and BeiGene, former Senior Medical Director at Takeda, Chief Clinical Integration Officer at Stony Brook Medicine, founding chair of biomedical informatics at SUNY, former Global Chief of Strategy at Medidata, and COO at P1 Pratia Oncology. All advisors joining since April 2026 invest a minimum of $25K in the SAFE. Skin in the game.
How Burna compares.
How is Burna different from the prior generation of clinical AI?
The previous generation of clinical AI in oncology failed on three architectural choices: broad scope, opaque reasoning, and autonomous decision-making. Burna is the opposite shape on every axis: narrow (CTCAE grading and multi-drug attribution only), transparent (citation on every grade), human-in-the-loop by design, and unified across clinical trials and postmarket on the same platform. The architectural shape is the lesson learned from what came before.
How is Burna different from Veeva Vault Safety, Argus, ArisGlobal, or Medidata Detect?
Those systems manage cases, workflows, reporting, and submissions after attribution has been made. Burna operates upstream of the attribution decision, helping the investigator and the safety team make the original grading and attribution call at the point of clinical encounter, while the source note, CTCAE criterion, regimen profile, and named algorithm are still visible together. Burna integrates with downstream PV systems; the two are complementary in a sponsor PV stack.
How is Burna different from Veeva Vault Safety AI Agents specifically?
Veeva Vault Safety AI Agents (launched April 2026) handle ICSR lifecycle management downstream of the attribution decision. Burna operates upstream of the attribution decision. The two are complementary. Burna integrates with Veeva Vault EDC and Veeva Vault Safety.
How is Burna different from clinical AI scribes (Abridge, Suki, Doximity GPT)?
Different category. Scribes generate clinical notes from ambient capture. Burna grades adverse events from existing clinical notes against the CTCAE ontology with citation discipline. The buyer is different (research operations, not clinical care). The regulatory frame is different (clinical decision support for trial safety, not clinical documentation).
Could an LLM wrapper match Burna’s results?
An LLM wrapper hallucinates. The Burna engine structurally cannot produce a grade outside the defined CTCAE set, cannot skip citation, and cannot contradict its own upstream findings. The output space is restricted by architecture, not by post-hoc filter. Twelve agents, cascading output boundaries, two patents filed.
Could someone build this in a weekend?
No. The pipeline spans twelve agent categories: term identification, category assignment, severity grade determination, multi-drug attribution, citation, and confidence calibration, with the constraint logic between them. Two patents filed on the architecture. The validation roadmap (three pre-registered studies, peer-reviewed target venues, council-lead co-authorship) is a separate moat that does not compress.
Will Veeva or Medidata add this as a feature?
They operate downstream of the attribution decision. Burna operates upstream. The white-space at the attribution layer (helping the investigator make the original grading and attribution call at the point of clinical encounter, with citation discipline and named-algorithm reasoning) is open. Burna integrates with Veeva Vault EDC, Veeva Vault Safety, Medidata Rave, and Oracle Clinical One. The architectures are complementary.
For the narrower questions.
This page covers the cross-cutting questions. Each audience page carries a narrower FAQ tuned to the work that audience does.
- Clinicians and coordinators →Workflow, language support, training feedback loop.
- Cancer centers →Pilot structure, integration timelines, EHR coverage, sponsor monitor implications.
- Pharma and biotech →Protocol Safe, MedDRA coverage, regulatory readiness, sovereignty.
- CROs →Query economics, white-label, Aria integration, tenant isolation.
- Patients and families →How the Patient Portal works, privacy, emergencies, language support.
- Investors →Round mechanics, comparables, exit path, advisor structure.
- IT and Security →Architecture brief, BAA, residency, SOC 2, EDSTP posture.
The conversation that fits.
For procurement and IT review, the 5-page Architecture and Implementation Brief is at security@burna.ai. For design partnership, partnerships@burna.ai. For investor diligence, nnenna@burna.ai. For press, press@burna.ai.
Book a 30-minute conversation →Not ready for a conversation?
Subscribe to the quarterly safety and data quality briefing. Validation readouts, platform updates, and conference timing only. No tracking beacons, no marketing email cadence.