Frequently Asked Questions

Start with the General model and a form targeting roughly 20–50 unique fields, then evaluate the extraction output together.

Share your target document set and we will provide tuned examples or prompt refinements as needed. Bank Statement extraction models currently require the dedicated Bank Statement form structure; broader form support is in progress.

Yes. We offer finely tuned workflows covering a broad range of SMB underwriting and B2B lending scenarios, including origination, underwriting, monitoring, broker introductions, and more.

Our ready-to-embed components deliver a complete review workflow for any document processed through the API.

• PDF Viewer with Citations – interactive overlays inside the PDF
• Citation Highlighting & Jumping – click a citation to highlight and scroll to the source region
• Bidirectional Navigation – jump from field citations back to the originating PDF location
• Data Display – render extracted data as multi-row tables and key–value pairs
• Inline Editing & Feedback – correct values in-place and open a feedback modal
• Embeddable – React today; Angular parity planned (see roadmap)

We maintain React/Angular-ready components so you can preview uploaded documents immediately.

Yes. The extraction pipeline considers all supplied documents when populating a submission.

For multi-year data, add a collection to your form schema, and the system automatically populates multiple rows (see the Income Statement form example). You can optionally include a userPrompt in the submission request (same field as the UI prompt) to emphasise instructions such as “extract all possible years.” The new Agentic model already follows nuanced directions well, so this hint is rarely required.

Yes. Citation metadata, including bounding regions, is returned from the /submissions/data endpoint within the documents block.

We are also adding direct citation deep links that load the widget with the citation pre-highlighted, and can expose an optional endpoint dedicated to citation references per document.

Choose the flow that matches your UX:

1. Deal ID + Form ID: load directly into submissions for that combination (recommended for PoC).
2. Deal ID: user selects which form to populate for the deal.
3. Form ID: user chooses the target deal for that form.
4. Submission ID: continue a specific in-progress submission.
5. Empty: allow the user to pick both deal and form.

Deals can be created or retrieved through the API before bootstrapping the widget. We are refining the embedded UX so the appropriate flow loads automatically once we understand your preferred configuration.

Use short-lived tokens minted server-side instead of a long-lived API key in the client. We will provide a lightweight “token vending” snippet (edge/serverless friendly) that exchanges OAuth client credentials for a single-use, short-TTL token consumed by the widget.

Alternatives include embedding via an iframe that relies on our hosted authentication, or implementing origin-bound token issuance with mechanisms such as DPoP if you need a purely front-end integration.

Yes. Our AI workflow agent guides you through building processes with the form builder, making it straightforward to match your lending workflow.

Getting started with Sea.dev takes days, not months. From kickoff to deployment, start building in one week, validate your agent in two weeks, and deploy to production immediately after. The timeline is fast without sacrificing quality.

Building a custom agent is straightforward. Choose chat, voice, or multi-modal, upload your financial documents (statements, underwriting policies, covenant thresholds), and Sea.dev crafts an agent tailored to your lending workflow.

Our agentic loop extracts data, validates accuracy, auto-heals errors, and learns from every interaction. The system becomes more reliable with each spread.

Clients typically reduce manual spreading from 35 hours to 2 hours while maintaining 99% extraction accuracy. Analysts review the results while our AI handles the bulk of the work with human-in-the-loop quality control.

Yes. AI-assisted form creation is available via the forms API. You can generate a draft schema and refine it using the standard form endpoints.

Incremental edits (adding groups, fields, or updating metadata) are supported through the enhanced forms CRUD endpoints, so you can manage the entire lifecycle programmatically.

1. Create a submission.
2. Upload documents via the extractions endpoint.
3. Poll the extraction status endpoint until processing completes.
4. Fetch merged results (including citations) from the submissions data endpoint.

Yes. The components are modular, and we are packaging them so you can compose exactly what you need. Modules include the form creation widget, submission & review widget, and a feedback/corrections workflow widget, each embeddable on its own or in combination.

Yes. We offer React components today and are delivering Angular parity so the same experiences can be embedded in Angular applications.

There is no limit on deals. Pricing and usage limits apply to submissions rather than deal count.

Sea.dev was designed to handle complex document workflows out of the box, including:

• Extracting PDFs, spreadsheets, and images into structured JSON enriched with citations.
• Categorising documents into nuanced, multi-label types, even across multi-page files with section-level splits.
• Orchestrating multiple forms in a single workflow to support sophisticated, multi-step processing.