● Now in private beta with 4 design-partner pharma teams
Medical writing, sourced by construction.
AI-assisted authoring for pharmaceutical medical affairs — where every claim is pinned to its evidence, every reporting standard is enforced, and every export is MLR-ready. From the first draft to journal submission, on one auditable platform.
Efficacy and safety of zelidravir plus dexamethasone in relapsed or refractory multiple myeloma
Abstract — Results
At a median follow-up of 17.2 months, median progression-free survival was 14.2 months (95% CI 12.1–16.7) with zelidravir + dexamethasone vs 6.8 months with dexamethasone alone (HR 0.41; P<0.0001)17. The overall response rate was 71.4% with zelidravir vs 49.8%1.
Investigators also observed early evidence of activity in patients who had previously received BCMA-directed therapy⚠ needs source.
Grade ≥3 treatment-emergent adverse events occurred in 62.1% of zelidravir-treated patients vs 41.0% in the comparator arm…
In private beta with medical-affairs teams across oncology, immunology and rare disease
VelorixTherapeutics
HalcyonBioPharma
NorthcrestOncology
LighthouseMedComms
AtriumSciences
SolwayHealth
The problem
Generic AI doesn't ship in pharma.
The bottleneck in medical communications isn't drafting — it's verifying every claim, conforming to a dense set of regulated standards, and shepherding the document through five-to-fifteen rounds of MLR review. Generic writing tools make the first step faster and the rest worse.
Today, a typical Phase 3 manuscript
~80 deliverables across the publication plan, 5–15 MLR rounds per asset, 8 systems in the writer's daily workflow.
01
High evidentiary burden
Every factual claim — every p-value, every response rate, every safety statement — must be supportable from a citable source. Writers spend more time hunting references and re-checking numbers than drafting.
02
Regulated by construction
Reporting standards (CONSORT, PRISMA, STROBE, ARRIVE) prescribe exactly what must be reported and in what order. Journal styles, promotional codes (ABPI, EFPIA, PhRMA), and ICMJE / GPP 2022 author rules layer on top.
03
MLR review is the bottleneck
A single manuscript may pass through 5–15 internal Medical/Legal/Regulatory review rounds before submission. Most of those rounds re-litigate the same handful of source-attribution and fair-balance questions.
04
Asymmetric compliance risk
A single off-label or unsupported claim slipping through can trigger an FDA warning letter, an ABPI complaint, or worse. Reasonable people respond with conservatism — and conservatism scales as slowness.
What's different
Three load-bearing differences from a generic writing assistant.
01 / Grounding
Source-grounded by construction.
Every sentence generated by the platform is anchored to a span in your source set — a CSR page, a SmPC paragraph, an indexed publication. Unanchored assertions are flagged as "needs source" and physically cannot be exported.
"Median PFS was 14.2 months (95% CI 12.1–16.7) with zelidravir vs 6.8 months1."
↓ retrieved span ↓
CSR AURORA-3 CSR v3.2 · p.1428, col. 2, ¶3
"Median progression-free survival was 14.2 months (95% CI 12.1–16.7) in the zelidravir + dexamethasone arm and 6.8 months (95% CI 5.4–8.1) in the dexamethasone-alone arm…"
02 / Compliance
Standards enforced as you write.
Off-label phrasing flagged against your label library. CONSORT, PRISMA and STROBE checklists generated automatically. ICMJE authorship, GPP 2022 disclosures and journal-style references handled as first-class data — never as a last-minute scramble.
✓CONSORT 2010 — 31 of 36 items completeauto
✓ICMJE author criteria — 7 of 7 meetauto
⚠Off-label population in §Methods · enrolmentblock
✓Fair-balance — safety adjacent to efficacy claimsauto
03 / Audit
Built for the MLR cycle.
Reviewers see, per claim, exactly which source spans were retrieved. Every AI invocation, every author decision and every reviewer signature is captured to an immutable, 21 CFR Part 11-grade audit trail — without anyone having to remember to log it.
12m ago
Maya accepted AI rewrite in §Abstract · Results
Claude 4.5 · grounded in r1, r7 · cost 0.34¢
52m ago
Marcus resolved claim c4 stat-mismatch
Changed P<0.001 → P<0.0001 to match source
2h ago
Dr. Lin approved §Methods
External author · guest link · e-signature
4d ago
MLR Round 1 complete · Jordan
18 redlines · 12 resolved · pushed to Veeva
Product tour
One platform from first draft to journal submission.
No more juggling Word, EndNote, PubMed, SharePoint, Veeva and three statistical-output PDFs in tabs. Every step happens against the same evidence set.
The drafting surface, redesigned around the citation.
Compose long-form medical content with a serif-typeset editor that treats claims as first-class objects. Each claim carries category (efficacy, safety, statistical, mechanism), status, and the source spans that ground it. Highlight any sentence to invoke a retrieval-augmented rewrite — drawing only from the sources you've selected.
Source-only generation. The AI cannot draft from its own training data; only from your CSRs, labels, indexed publications and data on file.
Cite-as-you-go. Inline footnotes pop a source preview with the exact paragraph quoted.
Multi-author co-editing. Real-time presence with named comments and tracked changes per author.
At a median follow-up of 17.2 months, median PFS was 14.2 months with zelidravir + dex vs 6.8 months with dex alone (HR 0.41; P<0.0001)17. The overall response rate was 71.4% vs 49.8%1.
Grade ≥3 treatment-emergent adverse events occurred in 62.1% of zelidravir-treated patients vs 41.0%1; the most common were neutropenia, thrombocytopenia, and infections.
Introduction ¶3
Investigators also observed early evidence of activity in patients who had previously received BCMA-directed therapy⚠ needs source.
✦Ask AI to draft, rewrite or find sources →r1r7
One screen for every claim in every asset.
The claim audit is a ledger of every assertion in the document — what it says, what category it falls in, what source it's anchored to, and what status it carries. Filter by status to triage off-label, needs-source, and statistical-mismatch claims first; one click jumps you to the exact PDF span.
Off-label detection. Claims compared against your product's current label library (SmPC/USPI per market).
Numeric consistency. Every numeric claim re-checked against the cited source value.
Fair-balance gaps. Efficacy without proximate safety, flagged for the content types that require it.
One-click navigation. From claim to source span to the exact line in the original PDF.
claim audit · MS-OM-0142
EfficacyOff-label
Patients were eligible to enrol regardless of prior transplant status, including newly diagnosed, transplant-eligible patients.
StatisticalStat mismatch
The primary endpoint was significant at P<0.001 in our analysis.
EfficacyNeeds source
Early evidence of activity in patients previously treated with BCMA-directed therapy.
EfficacyVerified
Median PFS was 14.2 months with zelidravir + dex vs 6.8 months…
C7 · off-label · blocked
"Patients were eligible to enrol regardless of prior transplant status, including newly diagnosed, transplant-eligible patients."
⚠ Off-label population
ZEMLIRA® USPI §5.3 — indicated only for patients with ≥1 prior line of therapy. Newly diagnosed, transplant-eligible use has not been established.
MLR review without the document-by-email tango.
Configurable Medical, Legal, Regulatory and Compliance review with role-aware SLAs. Each reviewer sees inline source attribution for every claim, leaves anchored comments, signs decisions with their corporate identity, and the decision pushes to your existing system of record. No more "what version is this?".
Veeva PromoMats / Vault MedComms two-way sync — read on import, push on decision.
Reviewer-ready PDF pack for legacy workflows: claims + spans annotated.
Round analytics per asset, per reviewer, per comment category.
21 CFR Part 11 e-signatures with reason-for-signature capture.
mlr · MS-OM-0142 · round 2
MLR · Round 2SLA 12hReviewer · Jordan Castelli
Efficacy and safety of zelidravir + dexamethasone in R/R MM
JC
Off-label population in §Methods. Refers to newly diagnosed, transplant-eligible patients — outside USPI §5.3. Must be removed or quarantined before MLR R3.
↳ §Methods · enrolment criteria
MR
Please clarify investigator-assessed vs BICR for the ORR claim. The abstract sentence is ambiguous as written.
↳ §Abstract · ORR
HL
Could we add the MRD subgroup forest plot here? I think the lay reviewer will want it.
↳ §Results · subgroups
Submission packs that pass the journal's validator on the first try.
One click assembles the exact package the target outlet requires — manuscript in the journal's file format, references in the journal's style, separately exported figure files at the required dpi, declarations, cover letter, completed reporting-standard checklist, and the ICMJE author forms. Every claim still carries its source. The audit log is appended.
Editorial Manager and ScholarOne direct submission with decision sync.
Congress packs for ASCO, EHA, ASH, IMS and many others — abstract structure, embargo and disclosure rules captured.
Plain-language summaries at configurable reading levels, with claim-to-source mapping preserved.
Translation via memoQ / Trados memories — reference numbering and claims maintained across languages.
12 weeks in, the numbers from our private beta look like this.
Pooled, anonymised, self-reported by design-partner medical-affairs and agency teams. Your mileage will vary by therapy area, asset mix and current baseline.
Time-to-first-draft
−51% vs baseline
Routine deliverables (abstracts, posters, slide decks) drafted in days, not weeks.
MLR rounds per asset
−36%
From an average of 2.8 rounds to 1.8 across observed manuscripts.
Citation-trail validator
100%
Of exports pass the platform's citation-trail check. Unresolved claims block export.
Writer satisfaction
+42 NPS
Among senior medical writers using the platform daily (n=23, beta cohort).
Who it's for
Built for the way pharma actually works.
Designed with writers, publications managers, medical directors, MLR reviewers and external KOL authors in the room. Not for any one role at a time.
MO
Maya · Senior Medical Writer
Agency · 8 years experience
I spend half my day chasing references and reformatting citations. medcomms-writer hands me a draft that already has its receipts.
EditorAI rewriteReferences
PA
Priya, PharmD · Publications Manager
Top-20 pharma · oncology
I run 8 manuscripts and 20 congress submissions a year. I need to see exactly where each one is, what's blocking it, and whose disclosure is missing.
DashboardPublication planAuthors
MR
Marcus, MD · Medical Director
Therapy-area lead
I'm the final sign-off. If a claim can show me its source span in one click, I can review three times the volume in the same hours.
Claim auditMLR reviewe-Sign
JC
Jordan, RAC · MLR Reviewer
Legal / Regulatory
My job is to catch off-label, unsupported and promotional drift. This platform catches the easy ones before they reach my queue.
MLR queueOff-labelAudit trail
HL
Dr. Lin · KOL Lead Author
External · UCSF
I don't want a new account or a new password. I want a secure guest link, the manuscript, and a comment box that anchors to the right sentence.
Guest linksCo-editingORCID
RT
Reena · IT & Compliance
Vendor security review
Show me your SOC 2, your DPA, your 21 CFR Part 11 evidence, and how you handle our data. Then we can talk seats.
SOC 2GAMP 5SSO/SCIM
Design partner
A.S.
Annika Sørensen
VP Medical Affairs, Halcyon BioPharma
The first MLR round on our zelidravir manuscript came back with fewer comments than the second round usually does. Half of the comments my reviewers used to leave are now resolved before the document leaves the writer's hands — because the writer can see the same flags I would have raised.
Compliance & security
Built to pass your security review, not to delay it.
Every assumption we made about authentication, audit trails, validation evidence and data residency comes from sitting with pharma IT, compliance and legal teams during scoping. We can hand over the documentation pack on day one.
21 CFR Part 11 / EU Annex 11
Tamper-evident audit trails · e-signatures with reason-of-signing
GAMP 5 (2nd ed.)
Validated SDLC · IQ/OQ/PQ summaries available under NDA
SOC 2 Type II
In flight · target H2 2026
HIPAA — BAA available
Enhanced controls for PHI in case reports
GDPR & UK GDPR
DPA · SCCs · DPIA template provided
Data residency
EU · US · UK · on-prem option for enterprise
SSO & SCIM
Okta · Azure AD / Entra ID · Ping · MFA required
No-training guarantee
Contractually enforced with every LLM provider
Hallucination controls, by design
Source-only retrieval. Post-hoc claim re-anchoring. Numeric-consistency checks. Off-label and content-safety classifiers before output reaches the user.
Model upgrades are opt-in per customer with 30 days' notice and a regression run against your pinned validation set.
Integrations
Fits the stack your team already uses.
We don't replace publication planning, MLR workflow systems or translation memories. We integrate with them — and behave well in the gaps between them.
Literature
PubMed · MEDLINE
Reference lookup and bibliographic enrichment via E-utilities. Crossref DOI resolution included.
Literature (license)
Embase · Cochrane
Broader biomedical search and systematic-review coverage where your institutional license permits.
Reference managers
EndNote · Zotero · Mendeley
Import RIS / BibTeX or set up live two-way sync. Per-organisation deduped library.
Publication planning
Datavision · PubsHub
Asset and publication-plan sync. Status reads back into Envision/ICON dashboards.
MLR · DAM
Veeva Vault PromoMats · MedComms
Read on import. Push approved versions and decisions back, with full version state preserved.
Submissions
Editorial Manager · ScholarOne
Submit directly. Track journal decisions and route responses against the same source set.
Plagiarism
iThenticate · Crossref
Similarity checks against the published literature and against your organisation's own prior assets.
Translation
memoQ · Trados
Use your existing translation memories. Claim-to-source mapping survives the language jump.
Identity
Okta · Azure AD · Ping
SAML 2.0, OIDC and SCIM 2.0 provisioning. MFA required across all customers.
Authoring
Microsoft 365 add-in
For Word-resistant authors and reviewers who prefer to stay in their habit.
Registries
ClinicalTrials.gov · CTIS
Pull trial registration metadata at asset creation. EU CTR and EudraCT supported.
Extensibility
REST API · Webhooks · SIEM
Wire your own dashboards. Stream audit events to Splunk or Microsoft Sentinel.
Pricing
Seats for the team. Add-ons for the throughput.
Per-seat subscription with metered AI generation and translation. Enterprise tier for single-tenant deployment, custom data residency and validation support.
AI generation, translation and plagiarism checks are metered. Most teams stay within included monthly volume; transparent per-unit pricing beyond.
FAQ
Things pharma teams always ask first.
No. Our agreement with every LLM provider includes a contractual no-training clause for customer content. Your source materials, prompts, generated outputs and retrieved spans are never used to train any model — frontier or otherwise. We can produce the relevant provider agreements as part of your security review.
Three layers. (1) Retrieval-augmented generation is the only way to produce text — the model is constrained to your selected source set and asked to refuse rather than freelance. (2) Every generated sentence is re-anchored to a source span by a separate verification pass; anchors that fail are surfaced as needs-source. (3) Numeric claims are independently checked against the cited source value before they ever appear in front of you. Unresolved claims block export.
No — and we don't intend to. PromoMats remains your system of record for MLR workflow and document management. medcomms-writer is the authoring and review surface that feeds it. Veeva integration is a P0 in our roadmap.
We develop under a validated SDLC and provide validation artefacts to enterprise customers under NDA — validation plan, traceability matrix, IQ/OQ/PQ summaries, change-control evidence and pinned-model regression suites. SOC 2 Type II audit is in flight; ISO/IEC 27001 follows within 24 months of GA.
EU, US and UK at launch. Single-tenant and on-prem deployments are available for enterprise customers with stricter residency requirements. All data is encrypted at rest (AES-256) and in transit (TLS 1.3). Customer-managed keys available in the Enterprise tier.
Yes. External authors receive a secure guest link with scoped permissions. They can read, comment, approve and e-sign without a provisioned seat. ORCID is supported for identity attribution. We strongly recommend MFA for guests; SSO-federated guest identity is on the roadmap.
Primary manuscripts, review articles, congress abstracts, posters, oral slide decks, plain-language summaries, standard response documents, advisory-board briefings, MSL scientific decks, peer-review responses and letters to the editor. EU CTR Annex V lay summaries are supported via the PLS content type.
A reasonable monthly volume of AI generation, retrieval and translation is included in each seat tier. Volume beyond that is metered transparently per generation, per token of translation, and per similarity check. Most teams stay within the included monthly volume. Customer-supplied LLM (Azure OpenAI, Bedrock, on-prem) is available on the Enterprise tier — you pay your own model bill.
Stop chasing references. Ship the manuscript.
Book a 30-minute working session with our team. We'll come with a draft of one of your live deliverables already loaded into the platform — and walk through the claim audit, MLR view and export pack against your actual source set.