Ben Dixon
Developer / EngineerAthens, Alabama
Custom websites up front.
Real business systems underneath.
I design and build branded websites, ecommerce experiences, dashboards, and internal tools that help businesses sell better and operate with less friction.
Proof of approach
One project, from first impression to operating layer.
Problem
A polished site is not enough if the lead, order, or content workflow falls apart behind it.
System
The demo ties a public experience to the private tools that keep sales and operations moving.
Result
You can inspect the customer path, the admin layer, and the handoff logic in one continuous walkthrough.
02 · Selected work
Real builds shaped by the business and the customer on the other side.
Each case study links to the full architecture, the decisions, and what shipped.
More technical breakdowns in the archive: storefront, admin, data model, and the workflow layer underneath.
Browse archive →03 · Flagship case study · Lead Intake System
Case study · portfolio-native reconstruction
Lead Intake System
A system design exercise focused on what happens after the form submission: structure, qualification, routing, follow-up, and human review. The product needed to behave like an operating layer, not like an email inbox with a nicer coat of paint.
At a glance
Internal tool / CRM workflow
Project type
Capture, qualify, route, operate, review
Focus
Product framing, interface design, system architecture
What I handled
Portfolio-native reconstruction
Status
01 · Capture
One form, shaped around the signals that actually matter.
Problem
Intake forms collapse into one of two failure modes. Either a single 'tell us about your project' text field that leaves the pipeline with nothing to work with, or a fifteen-question form heavy enough that prospects abandon before the contact block. Neither gives the downstream system enough structure to score or route reliably.
Decision
Capture four required fields that actually predict how to handle a lead: services (multi-select), budget bracket, timeline, and a short project brief. Plus contact and company. Everything else stays optional. The schema is enforced on the client and the server so the pipeline never receives a half-formed record.
Result
Every captured lead arrives with enough shape for scoring, routing, and the review surface to treat it as a first-class record instead of raw text.
/inquiry
/admin/leads/[id]
02 · Qualify
AI scoring that fails safely.
Problem
Pure AI scoring has two obvious risks. It can't be the single point of failure for qualification; an upstream API outage shouldn't pause inbound work. And a black-box score operators can't explain is a score operators won't trust.
Decision
Score every lead through GPT-4 with structured output: score, category, reasoning, urgency, suggested reply. Behind it, a deterministic scorer uses the same signals to produce a comparable number when the API is unreachable. Every result logs its source. When the model's confidence sits below threshold (Wren's missing budget, Reza's sub-floor budget plus referral, Adi's compressed timeline), the lead routes to manual review with the reason flagged, not a confident guess.
Result
Qualification is never blocked by an upstream dependency, and the operator can always see why a lead landed in a given tier before accepting or overriding it. Scoring weights and the signal taxonomy live in Settings, so the AI is tuned, not magic.
03 · Review
The AI drafts. The operator sends.
Problem
AI auto-responses that fire after a black-box confidence check end one of two ways: confidently wrong (sent something tone-deaf under your brand), or paralyzingly cautious (everything queues forever and the prospect goes cold). Neither is good UX for operator or prospect.
Decision
When the model's confidence holds, the response goes out immediately with full audit. When it doesn't, the AI still drafts the reply, but it lands in the operator's Review & send queue with the reason flagged: missing budget signal, sub-floor budget plus referral, compressed timeline below typical lead times. Operator reviews, edits inline, approves or skips. The drawer opens over the dashboard so the inbox context never disappears.
Result
Speed where it earns it, oversight where it matters. The operator never types from scratch on a fallback case, and the AI never sends something it shouldn't.
/admin/inbox/review-and-send
/admin/activity
04 · Ledger
Every action shows up, with reasoning.
Problem
Black-box AI workflows look magical for two demos and become legal liability by month three. If you can't explain why a lead got auto-routed to lost, the system is worse than a spreadsheet.
Decision
Every AI and human action joins one chronological ledger: auto-classified, draft generated, status changed, reply sent, follow-up queued. Each entry carries the reasoning the AI used, not just the label it produced. Filterable by AI or User. Visible across all leads.
Result
The AI is a system of record, not a labeler. Operators (and visitors) can audit any decision back to its inputs. The same NotificationCard renders three ways (bell, toast, panel), so a flagged action lands wherever the operator is.
05 · Try it
The whole surface, interactive.
Five panels, no sidebar; the dashboard is the workspace, not a navigation tree. Switch tabs at the top of the frame, click any lead row to open the drawer, and poke around. Everything is a static recreation; no backend needed to tour the UI.
bendixon.dev/work/lead-intake-system
STATIC
04 · Get started
Pick what fits. Every engagement starts with a short conversation. Most combine two or three of these.
Booking Jun–Aug 2026 · CT
01
Website
Branded, fast, built to sell. Custom design, a CMS your team can actually use, performance that passes the audit.
Brand & designCustom buildCMS setupAnalytics + SEO
View
02
Ecommerce
Shopify or custom, storefront through to back office. The one customers love and the one your team does not dread.
StorefrontCheckoutSubscriptionsWholesale / B2B
03
Internal tools
Dashboards, admin panels, intake systems. The surfaces your team lives in, as considered as the ones your customers see.
DashboardsAdmin UIsIntegrationsRoles & auth
04
Automation
The quiet plumbing. Orders to fulfillment, leads to CRM, forms to Slack. Removes the work that has to happen every morning.
n8n / MakeWebhooksAPI integrationsMonitoring
→
Start a website
Short intake · reviewed within 1 business day · free 30-min call
Engagement flow · same for every service
1
Intro call
A short, free call to see whether the fit is there. If it is, you get a proposal within a week.
2
Discovery
One to two weeks mapping what exists, what is missing, and what we are actually solving for.
3
Design & build
Weekly demos, Loom videos between them. You see it as it comes together.
4
Ship & support
Launch, then 30 days of included support. Retainers available after.