Scoop

AI drafting and reconciliation form the core of IND Sync. The software classifies uploaded scientific and manufacturing materials, maps their contents into a submission structure, and generates regulatory narratives. It then checks for inconsistent values and revises sections when source material changes. These learned document functions distinguish the service from a repository that merely stores or versions files.

Human editing, comments, and approval remain important, but the product's principal value depends on machine interpretation and generation. Its focus is the ongoing preparation and maintenance of related regulatory documents rather than a general chat feature added to an existing records system.

Users can edit and comment on generated drafts, and qualified staff must review AI output before relying on it. The service therefore assigns a real review role rather than presenting an automatic submission authority. At the same time, the homepage promises that filings never lose synchronization with the latest results. That absolute statement is not qualified by the separate estimate that drafts are 70% to 80% ready for review.

A missed source update or incorrect dependency could still leave an affected section unchanged. Detection thresholds, stop conditions, and recovery procedures for that failure are unspecified, leaving the completeness promise broader than the described controls.

Google Cloud Vertex AI and Document AI are named for AI, extraction, and document-processing requests. This identifies actual services involved in processing customer material, rather than relying on the founders' employment history to imply a supplier. The workflow also makes clear that the system classifies inputs, maps them to regulatory sections, and generates or revises text.

Individual foundation model families, deployed versions, training datasets, and criteria for switching models remain unspecified. Vertex AI is a service platform capable of hosting different models, so its name alone does not identify the exact model responsible for a particular draft or update.

The service policy identifies Google Cloud Vertex AI and Document AI for AI and extraction, Vercel for application hosting, and Supabase for database and authentication functions. It separately names Framer for the public website, Resend and Twilio SendGrid for email, and Cal.com for demonstration scheduling. These distinctions help separate providers receiving clinical development content from those supporting marketing or communications.

The remaining uncertainty is the complete model route: specific downstream models, retention periods, training permissions, and the data sent to each inference service are not fully enumerated. A provider list therefore supplies meaningful visibility without a complete processing contract map.

Initial drafts are described as 70% to 80% ready for regulatory review, alongside a promise of much faster IND preparation. These figures concern writing readiness and elapsed preparation, not trial outcomes or clinical accuracy. A named customer, document sample, assessment method, reviewer qualifications, and measurement period are not supplied, so the readiness estimate cannot be reproduced or compared with a manual baseline.

Sentence citations and inconsistency flags describe how the product works, rather than demonstrating how often it succeeds. The evidence currently supports an intended workflow and an early performance claim, with substantial uncertainty about actual review effort.

Authorized AI requests send relevant customer content to contracted model and document-processing services. The service policy restricts use for advertising and sale, while return or deletion of customer content follows the customer agreement and lawful instructions. Retention also considers backups, incident response, and legal preservation, without fixed windows. Explicit model training restrictions and a defined deidentification method remain unclear.

Qualified review is required because generated content can be incomplete or inaccurate; citations and inconsistency flags support that review. Model-specific hallucination tests, stopping thresholds, and a safety incident process for erroneous regulatory content are not established by those general safeguards.

The August 2026 privacy policy expressly covers hosted software and customer content, including documents, records, prompts, and generated outputs. Business customers control that content, while Scoop acts as their processor or service provider under the customer agreement. This provides a substantive description of data roles. It does not establish HIPAA business associate status, and a readable BAA with covered products and subcontractor obligations is unavailable.

An agreement to process documents on a customer's behalf is not by itself a HIPAA commitment. The product's regulatory focus also does not mean that uploaded research materials necessarily exclude identifiable health information.

The service policy describes access controls, tenant isolation, encryption during transmission, logging, and restricted service credentials. Supabase handles authentication credentials, and Scoop says it does not receive passwords in readable form. A contact route exists for suspected account compromise. These are useful control descriptions, but no independent audit, named certification, or penetration-test report establishes an externally assessed assurance level for the current service.

IND Sync prepares and maintains documentation used in investigational drug submissions. Its functions include drafting, cross-referencing, formatting, and revising sections as scientific or manufacturing information changes. Qualified regulatory professionals retain responsibility for reviewing the work. This is an administrative research workflow, without a claim that the software diagnoses patients, selects treatment, or holds medical device clearance.

Preparing material for an FDA submission does not imply authorization of the software by that agency. Acceptance of a sponsor's filing would concern the particular investigational program, while the accuracy and adequacy of generated text still depend on the reviewed submission.

Source citations and qualified human review help users check individual passages. They do not define a formal model governance program. A named evaluation framework, release approval process, customer review artifact, or comparative assessment across patient groups is not established for IND Sync. This limits visibility into how systematic drafting errors or uneven representation in source material are detected across successive model changes.

Sentence citations, direct editing, and comments give regulatory staff a practical way to challenge and correct a draft before use. The privacy policy explicitly requires qualified review. Contractual accuracy warranties, output correction obligations, and remedies for erroneous submitted material remain unspecified because the customer service agreement controls the purchased product. The website's broad disclaimers and content license concern a different scope and do not establish the service's negotiated liability terms.

Veeva Vault is named as an integration target for the regulatory document workflow. The core process otherwise begins with uploaded study reports, manufacturing packages, and supporting files. Exchange direction, supported interfaces, version compatibility, and an implemented customer connection remain unspecified. A document-management integration claim therefore does not establish production exchange with a patient EHR or automatic writeback of approved submission material.

The product is a hosted service using Vercel for application hosting, Supabase for authentication and database functions, and Google Cloud for storage, document processing, and AI. Tenant isolation is among the stated security controls. Data may be processed in the United States and other countries where providers operate, with contractual transfer safeguards where required. A committed region, backup location, inference route, and customer-selected residency option remain unspecified.

Customers obtain access to a hosted software service under a separate customer agreement. The buying route begins with a demonstration request, and the website distinguishes its marketing terms from the agreement governing the purchased service. A public subscription amount, per-user or per-submission billing unit, document allowance, implementation charge, and minimum commitment are not specified.

Consequently, neither the advertised drafting speed nor the scope of IND maintenance establishes an initial budget. The product has a commercial service model, but the price and included capacity depend on contracting rather than a self-service plan that buyers can size in advance.

The intended users are biotechnology regulatory affairs teams, CROs, and consultants preparing investigational drug submissions. Inputs include scientific study reports and chemistry, manufacturing, and controls packages, making the initial workflow specific to development documentation rather than routine clinical care. Maintenance extends to new study results and manufacturing changes that affect an existing submission.

Small and large organizations are addressed, but named customer deployments and measured results across therapeutic areas remain unspecified. The range of document types indicates useful functional breadth within IND work; it does not establish equivalent performance for every disease area, sponsor size, or international submission pathway.