Meridian Capital
Editorial advisory website with an AI onboarding agent that reads client documents and waits for adviser sign-off.
Meridian is a concept for an independent advisory firm — the kind of business that wins clients in a room, then loses them to a website that looks like every other firm in the sector. The brief we set ourselves was to build trust without a single trust-signal cliché.
- AI pattern
- Document AI · human-in-the-loop
- AI component
- Onboarding document agent
- Typical timeline
- ~7 weeks
- Maps to
- Website Professional + AI Automation
Meridian Capital is a fictional company — this is the build plan we execute for a business like it.
Try it live
Use it, don't just read about it.
Watch an onboarding pack become a verified CRM record: extraction with confidence scores, cross-document checks, and adviser sign-off before sync.
- 01Hover a field to see where it came from on the page
- 02Confirm the smudged date of birth
- 03Open the joint review and resolve the address mismatch
- 04Approve and watch the CRM sync
Scripted and deterministic so it runs anywhere. In production the same interface runs on live models, your data and your systems.
01The problem
What's really in the way.
Financial sites converge on the same solutions: aerial cityscapes, stock portraits, and a wall of disclosure text nobody reads. The result is that every firm looks equally credible, which means none of them do.
- 01
Trust has to be constructed from structure and language, not imagery.
- 02
Regulatory disclosure is unavoidable and usually destroys the page's composition.
- 03
The audience skews older, which makes type size, contrast and keyboard access load-bearing rather than optional.
02What we'd build
The scope, plainly.
A calm, document-like website for an independent advisory firm, plus a document agent that reads each new client's onboarding pack — fact-find, statements, ID — extracts the fields, cross-checks them, and hands the adviser a verified file to approve before anything reaches the CRM.
Editorial website
Server-rendered pages for approach, advisers, fees and disclosure, built on a reusable type and hairline system.
Document extraction
Classifies each file, extracts structured fields with a confidence score, and links every value back to the exact place on the page it came from.
Cross-checks
Rules compare documents against each other — names, addresses, dates of birth, income against statements — and flag what doesn't line up.
Adviser sign-off
Low-confidence fields and failed checks need a human decision. Nothing syncs to the CRM until the adviser approves, and every decision is logged.
03Architecture
How the pieces fit.
Models are the swappable part. The tools, rules, logging and tests around them are what make a system dependable.
- IntakeL1
- Secure upload portal
- Email-in for adviser inboxes
- Virus scan + file typing
- AgentL2
- Vision-LLM extraction to a typed schema
- Per-field confidence + source region
- Cross-document rule engine
- PeopleL3
- Review queue for low confidence
- Accept / request-info decisions
- Client info-request drafts
- SystemsL4
- CRM upsert + document attach
- Suitability task creation
- Immutable audit log
- Next.js
- TypeScript
- OpenAI / Anthropic vision models
- Structured outputs (JSON schema)
- n8n / custom workers
- Postgres
- CRM API
- Vercel
04Screens
The interface, working.
Captured from the live demo in a headless browser — not mockups.
05Deliverables
What you'd own at the end.
- Design system and 6–8 page website
- Onboarding document agent for fact-finds, statements and ID
- Cross-check rules written with the firm's compliance lead
- Adviser review queue with approve / request-info actions
- CRM integration: client record, documents and review task
- Evaluation set of sample packs, run before every release
- Repository, documentation and handover call
06Typical timeline
Shipped in ~7 weeks.
- Discovery & content mapPositioning, page list, approved sources
- Design system & pagesEditorial grid, templates, disclosure treatment
- Build websiteStatic rendering, accessibility, SEO
- Document agent & evalsExtraction schema, checks, review queue
- Review & launchCompliance review, hand-off, handover
Schedule fixed in writing at kickoff.
07Approach
Decisions we'd take first.
Strategy
We treated the site as a document rather than a brochure — something with the authority of print, where hierarchy does the persuading.
- Lead with the positionThe homepage opens with what the firm believes about managing money, stated plainly, instead of a generic welcome.
- Disclosure as designRegulatory copy is given a deliberate, legible treatment in the page rhythm rather than being hidden in six-point grey.
- People before processAdvisers are introduced early, because in advisory work the relationship is the product.
Design
A tight editorial system: one serif for argument, one grotesque for interface, generous measure and a deep green ground that reads as considered rather than corporate.
- Two-column asymmetryA persistent narrow column carries labels, indices and disclosure, keeping the main measure clean.
- Rules, not boxesContent is separated with hairlines instead of cards, which keeps long-form pages calm at any density.
- Restrained paletteOne ground, one ink, one accent used only for actions — so the accent always means the same thing.
Engineering
Built as a static, server-rendered site: no client-side data fetching on any public page, and JavaScript reserved for navigation and the enquiry form.
- Static by defaultEvery public route pre-rendered, so time to first byte is a CDN lookup.
- Type-first loadingVariable fonts subset and preloaded, with metric-matched fallbacks so the headline never reflows.
- Accessible by constructionAA contrast throughout, a visible focus system, and a keyboard path through every interaction.
08Intent
What it's designed to do.
Success criteria are agreed and measured from launch.
- Build trust through structure and plain language rather than stock imagery.
- Turn a pile of onboarding PDFs into a verified client record in minutes, not an afternoon.
- Give advisers a cleaner first meeting, because the basics are already checked.
- Keep a person accountable for every record that reaches the CRM.
Your project
Want a plan like this for your business?
Book a 30-minute call. If it's a fit, we'll send a written scope and fixed price, or suggest starting with a $500 pilot.
Or email hello@xaevora.com · reply within one working day