Skip to main content
NELLA Labs

NELLA Labs platform

Building the platform this site runs on

The public site, the Project Architect, the CRM and the client portal are one system on one data model, so the story a buyer reads on the marketing site is the same object the delivery team works in.

Client
NELLA Labs — internal platform
Status
In active development
Year
2026
Sector
Professional services

This is a NELLA Labs product or platform, not an external commission. Nothing here implies a client engagement.

Context

This platform is our own. We are publishing it as work because the most honest demonstration of how we build is a system you can use rather than a description of one.

Challenge and constraints

  • A marketing site and an operational system usually diverge immediately, because they are built by different people from different sources.
  • An enquiry that arrives as a paragraph of free text gives neither side enough to have a useful first conversation.
  • Client data in a portal must be isolated absolutely, including through search, exports, realtime channels and error messages.
  • AI can genuinely help structure a rough idea, but it must never quietly change what a client said or issue a price.

Discovery

What we found, and what it changed

Research is only worth doing if it changes a decision. Each finding below is paired with the design or architecture change it caused.

Finding

The delivery lifecycle, the CRM pipeline and the portal phase labels were three different vocabularies for the same process in most organisations we have worked with.

What changed

The lifecycle became a single typed object that the public page, the pipeline mapping and the portal all read from, so they cannot drift.

Finding

Long forms lose people. The abandonment is concentrated at questions the person cannot answer rather than questions they do not want to answer.

What changed

Every question in the Project Architect offers "I don’t know — advise me", which records an open question rather than forcing a guess.

Finding

Published pricing is either uselessly vague or misleadingly precise.

What changed

Bands are published with their inclusions, exclusions and the variables that move them, and every figure is marked indicative and requiring human confirmation.

Solution

What we built

  • One route taxonomy, one design token system and one permission model across public, client and staff surfaces.
  • A versioned wizard definition with a branching engine, so the exact question set a client answered is retained with their submission.
  • Anonymous start with autosave, then magic-link save-and-resume — no account required before the work begins.
  • Row-level security in Postgres as the isolation boundary, with positive and negative tests per role.
  • AI assistance that proposes editable artefacts and records model, prompt version, inputs, outputs, cost and human overrides for every run.

Journeys

How people move through it

Visitor to structured opportunity

Start anonymously, answer branching questions with autosave, review an editable Project Canvas, and submit — creating an opportunity and a workspace.

Staff qualification

The opportunity arrives with a canvas, two separate scores, a routing decision and an SLA, rather than as an email.

Client workspace

The same submission becomes the client’s project home, with the brief, artefacts, approvals and messages attached to it.

Architecture

Decisions and the reasoning

These are the choices that would have been expensive to reverse. Each one is recorded as an architecture decision record in the project.

One data model across marketing, wizard, CRM and portal.

The divergence between what is sold and what is delivered is a data problem before it is a process problem.

Row-level security rather than application-layer authorisation alone.

A missed check in a route handler should not be able to become a cross-tenant leak.

Versioned wizard definitions with answers retained against the version.

A submission has to remain interpretable after the question set changes, or historical opportunities become unreadable.

AI behind a gateway with per-run records.

Cost control, provider substitution and the ability to explain an AI-assisted artefact all require the same instrumentation.

Security, accessibility and quality

Isolation
Positive and negative access tests per role across tables, storage paths and realtime channels.
Accessibility
WCAG 2.2 AA target with keyboard and screen-reader review of the wizard and portal journeys.
Performance
Static generation for public content; no WebGL or heavy motion on the critical path.
Honesty
A known-limitations register published with the build rather than maintained privately.

Timeline

  1. ShapeIn progressInformation architecture, content model, data model and permission model.
  2. BuildIn progressPublic site, Project Architect, schema and portal delivered in sequence.

Team shape

  • Product orchestration
  • Design system
  • Content and editorial
  • Platform engineering
  • Security review

Technologies

  • Next.js
  • TypeScript
  • Tailwind
  • Supabase
  • Postgres
  • Row-level security
  • Vercel

Outcomes

What we can actually substantiate

Each statement is marked either verified — meaning it is demonstrable in the system — or design intent, meaning it is what the design sets out to achieve and has not yet been measured in production.

  • VerifiedThe delivery lifecycle is one object shared by the public site, the CRM pipeline and the portal.
  • VerifiedEvery published price is marked indicative and requires human confirmation before it becomes an offer.
  • VerifiedUnverified business claims — addresses, availability, metrics — are labelled rather than published as fact.
  • VerifiedClient project data is isolated by row-level security with automated negative tests.

Next stage

What comes next

  • AI-assisted canvas generation with full run instrumentation
  • Proposal and estimate versioning with client viewing activity
  • Support, service reporting and renewal workflows

Related

Keep reading

Service

Create

Take a new product from idea to a first release people actually use.

Service

Automate

Apply AI and automation where they measurably pay for themselves.

Work

Building verification into the centre of a learning marketplace

A tutoring marketplace where an educator profile is invisible to families until identity and credentials have been checked, and where sessions accumulate into curriculum-level progress rather than billable hours.

Work

Treating identity verification as orchestration rather than integration

Verification modelled as a configurable, resumable sequence of steps with an append-only evidence trail, so adding a market or replacing a provider is a configuration change rather than an engineering programme.

Next step

Have a problem shaped like this one?

Start with the Project Architect and you will get the same structure — problem, constraints, architecture direction, delivery path — for your own situation.