Custom AI Operations and CRM Platform
12 production stages, 11 Supabase tables - replacing WhatsApp and spreadsheets for a bespoke jewellery operation.
TL;DR
A multi-level access portal platform for Adamas Studio, a bespoke jewellery operation spanning the UK and US, replacing fragmented WhatsApp and email workflows with a structured RFQ-to-delivery system. The platform includes an AI diamond scoring engine, repeat-customer preference learning, AI-assisted CAD and image generation, finance forecasting, and market pricing intelligence - alongside an eleven-table Supabase data model, a 12-stage production pipeline, and integrated Stripe, Sendgrid, Twilio, and blockchain diamond provenance tracking.
The brief
What did the client need?
Adamas Studio runs a custom jewellery operation across the UK and US. Every piece is bespoke: customer brief, diamond sourcing, vendor tendering, CAD design, production, quality control, delivery. Seven stages, multiple external production partners, high-value purchases where the customer needs to trust the process as much as the outcome.
Before the portal, all of that ran on WhatsApp threads, email chains, spreadsheets, and verbal handoffs. There was no centralized view of where any order stood. Vendors had no structured place to submit work. Nothing was audit-trailed. Customers couldn't see progress. For a business where a single piece might take twelve weeks and involve four different production partners, that wasn't a workflow. It was a memory sport.
The constraints
What made this hard?
The platform had to serve three completely different user types simultaneously: customers buying a high-value bespoke piece for the first time, production vendors executing specific parts of an order, and admin managing everything in between. Each portal had to be scoped so users only ever saw what was relevant to them. Merging those journeys, or over-exposing any one of them, would break trust in different directions.
The production pipeline ran to 12 distinct stages, from CAD Submitted through to Complete, with a parallel requirement to translate those into simpler, customer-facing equivalents. Vendors needed granular operational status tracking; customers needed confidence, not a production schedule. Those are different things, and they had to coexist in the same system.
Diamond provenance added a layer most platforms don't have to think about: blockchain tracking of each stone from creation through cutting to end manufacturer to the customer's hands, integrated into the broader inventory system.
The approach
How did Tincture frame the problem?
The structural insight was that a three-sided operation with fundamentally different user needs couldn't be served by a single interface. Customers, vendors, and admin had different information requirements, different trust thresholds, and different relationships with order status. Each portal had to be scoped to its user.
The other load-bearing architectural decision was the two-tier status system: 12 granular internal stages for operations, mapped to simplified customer-facing equivalents running in parallel. Vendors needed to know which stage of production they were in; customers needed to know their order was progressing. Collapsing those two needs into one status set would have broken one or the other.
The build
What was shipped?
I designed, spec'd, and built the entire platform from scratch.
The architecture is a three-portal system on Supabase, with eleven interconnected tables: Customers, RFQs, Quotes, Orders, Tasks, Task Subtasks, Vendors, Vendor RFQs, Order Status History, Email Logs, and System Settings. Every status change is audit-trailed.
The Customer Portal walks buyers through diamond sourcing with side-by-side certificate comparisons and an AI scoring engine that ranks each stone against the customer's stated specification - cut, carat, colour, clarity, budget. Repeat-customer preference learning refines those rankings across orders. Design iteration uses AI-assisted image and CAD generation so customers can explore visual concepts before anything goes to production. CAD review includes annotation and iteration tracking; payment runs through Stripe.
The Vendor Portal gives production partners a scoped view of their assigned work only: specifications, deadlines, production status updates, CAD uploads for admin review before customer visibility, and invoice submission directly through the platform. The chaos of scattered email communication replaced with a system that sets clear expectations and tracks everything.
The Admin Portal is the operational nerve centre: RFQ intake, vendor quote comparison, configurable markup calculations for diamonds and crafting, Stripe payment link generation, and conversion of paid quotes into production orders with milestone timelines. AI-powered finance forecasting surfaces cashflow predictions and margin analysis across active orders, and market pricing intelligence draws on live diamond market data to flag when sourcing costs drift against forecast.
Communications run through Sendgrid and Twilio with template-based messaging, channel preference tracking per customer, and automated notifications tied to status changes. Every communication is logged against the relevant order.
The outcome
What were the results?
A launch-ready platform replacing the full fragmented operations stack - WhatsApp, email, spreadsheets, verbal handoffs - with a structured, role-based system managing every stage from RFQ to delivery. Three complete user journeys. Nine core systems. An eleven-table relational data model. A full PRD built to be deployed incrementally, delivering value from the first module rather than requiring everything to land at once.
The admin team went from holding the entire order state in their heads to having a single operational view. Vendors went from waiting on emails to logging in and seeing exactly what was expected of them.
What it took
What tools and methods were used?
Supabase (relational database, role-based auth, real-time subscriptions), Stripe (payment processing and payment link generation), Sendgrid (transactional email), Twilio (WhatsApp notifications), blockchain provenance tracking for diamond chain of custody, React (three separate portal frontends with distinct routing and access control), role-based access control enforced at database level. AI layer: diamond scoring engine and repeat-customer preference learning, AI-assisted CAD and image generation for design iteration, finance forecasting and cashflow prediction, live market pricing intelligence.
The takeaway
What's the transferable principle?
Bespoke operations have a specific problem: the complexity that makes the business special is exactly the complexity that makes off-the-shelf tools useless. When every order is different, a CRM built for repeat transactions won't hold the shape of your workflow. You can bolt on tools at every stage, but the seams always show, and the order state lives in someone's WhatsApp.
The platform has to be built for the actual process. Not approximated from something close.
Frequently asked questions
More like this
12 agents, ~10 issues/day, ~2 hours saved daily, cost = Claude subscription + ~$40/month VM
A 12 agent ↔ Linear ↔ Notion setup
The Operating layer was wired to Linear and Notion, paired via MCP, so each specialist agent could ship from Linear and mirror state into Notion against its per-persona write contract, with identity that survives across sessions. Stand-up took about 2.5 working days of focused build, over a week of elapsed time. Steady state: 12 agents working roughly 10 issues or tasks a day, about 2 hours of time saved daily plus the context-rebuilding tax that doesn't show up on a clock. The whole thing runs on a Hetzner VM in Helsinki, so agents pick up Linear comments and project emails whether the laptop is open or not. Cost is the Claude subscription plus about $40 a month for the VM.
~$70k of private-client revenue pre-launch
A 1m+ SKU marketplace from concept to launch
Tincture led Product and Operations as co-founder of Adamas Studio's 1m+ SKU lab-diamond marketplace, building the entire commercial and operational backbone from zero across multiple PRD iterations. The marketplace consolidates two vendor APIs into a single live-inventory results page, layered with a Custom GPT CAD generator, an Ideal Diamond finder, and a Reddit-driven market intelligence engine. Roughly $70k of private-client revenue shipped before the marketplace went public, on infrastructure designed for scale rather than launch.
A mobile-first GCSE revival tool with ten subject tracks.
GCSE Revival: A mobile-first revision tool
We built a mobile-first GCSE revision and reintegration tool from scratch for a fifteen-year-old returning to school after extended hospitalization, with ten subject tracks tailored to specific exam boards, a confidence-first design philosophy (week one is reintegration, not content), gamified XP and mastery progression, timed exam-condition exercises, and a PIN-protected parent dashboard tracking session progress and confidence trends.
Facing a similar problem?
Every engagement starts with a diagnostic. No pitch decks, no proposals. Just the gaps, named plainly.