Defined the product model
I reframed client requests as configurable workflows instead of isolated AI features, giving the platform a repeatable structure for new banking processes.
Synthex · 2024 - 2026
AI workflows for banks.
Led design from a name and a demo to a funded round.
Synthex
AI workflows for banks and advisors
From a name and a demo to a funded product.
When I joined, there was no product and no production code. I built the workflow product model — processes, flows, and nodes — and shipped React prototypes with Cursor and Claude that became the investor artefacts for the first round.
Ownership
Product and design from day one
Product model, first interfaces, design system, and path to scale.
Team
Four people, full loop
Product, design, and engineering in one tight cycle — no handoff theatre.
My edge
Clients shaped agents and UX
Ongoing contact with banks and advisors directly shaped agent architecture and platform UX.
Context
Used by banks and advisory firms for high-stakes process work.
Synthex consolidates data from legacy systems, PDFs, and calls into configurable workflows for process analysis, suitability, KYC/FI, and client communication. The product had to make complex regulated work faster without making advisors feel they had lost control.
01Legacy data
Inputs from systems, documents, calls, and client records.
02Regulated tasks
Suitability, KYC/FI, compliance checks, and reports.
03Configurable flows
Processes modeled as reusable nodes instead of one-off features.
04Advisor control
Human review stays inside the loop before client-facing output.
My contribution
I turned an AI demo into a product system banks could understand.
My value was not only designing screens. I created the product language, the first UX model, the design system, and the prototype logic that helped the team explain the opportunity to investors and clients.
Made AI feel controlled
I designed validation, review, and handoff moments so advisors could trust automation without losing control of regulated work.
Designed close to code
I built prototypes in React to test interaction decisions in working code — AI-assisted speed without dropping the quality bar regulated work demands.
From zero to product
Name, demo, first interface, investment.
When I joined, Synthex had no production product: only a demo where a call could auto-fill an advice form. The first job was defining the product shape, the interface language, and the structure that made the opportunity credible to clients and investors.
01 · Problem
No product, just a problem
I helped define what we were solving, for whom, and why it mattered.
02 · Design
First interfaces built to scale
I created the first product structure, components, and scalable workflow model.
03 · Outcome
Design-led first round
Working React interfaces made the case to investors before any production code existed.
The charts
The manual work had to become visible first.
Advisors were spending 30+ hours per user per week analyzing documents and calls, checking transactions, building Excel sheets, reports, and presentations. That time came directly out of time with clients.
Workflow model
Processes, flows, nodes, validation.
The core product bet was that new client capabilities should be configured as workflows, not rebuilt as separate features.
That decision made the product easier to sell, easier to explain in demos, and easier to scale across client-specific banking processes.
Client data
Document analysis
Suitability check
Draft report
Human review
AI model
Agents beyond the chat box.
Instead of confining AI to chat, we placed agents, sub-agents, and skills inside the workflow. The workflow gives AI structure and boundaries while each node handles a specific job: reading, checking, drafting, reviewing, or preparing output.
That made AI feel like part of a regulated process, not a magic text box sitting beside the product.
Agent
Sub-agent
Skill
Node output
Human approval
Trust and validation
In fintech, an AI output nobody trusts is worthless.
Agent outputs were checked by validation steps built into the workflow. Automated steps kept the advisor in control: for example, pre-drafted emails could be reviewed and sent in one click instead of being sent automatically.
Validation
Benchmarks before shipping
Each workflow could compare model outputs before selecting the best fit.
Review
Human in the loop
Critical outputs create review moments instead of invisible automation.
Control
Automation with approval
The product speeds up work while preserving regulated decision ownership.
Research and delivery
I partnered directly with banks and advisors to define agents and UX.
I worked with banks and advisors throughout — not one-off demos, but ongoing contact mapping their real workflows. I translated that contact into the workflow model, agent behavior, and every major UX decision.
Map real client workflows
Heavy contact with banks and advisors to understand how they actually work — not assumptions.
Shape agents from process
Each client process revealed where agents, sub-agents, and skills needed to sit in the workflow.
Refine UX from feedback
Client sessions drove platform UX — validation, review moments, and control patterns.
Design system and design-to-code
A platform needs foundations from day one.
Treating Synthex as a platform meant building a design system from the start, so new flows could ship without redrawing the whole product.
I used Cursor and Claude as a playground to move from wireframe to high fidelity, validate interaction decisions, and workshop the system with developers.
01Foundations
Platform-scale design language from day one — not screen-by-screen craft.
02Scale
New banking flows reuse existing patterns instead of one-off UI rebuilds.
03Trust
Validation, review, and handoff rules built into how the system behaves.
04Delivery
React prototypes turned design decisions into working product input.
Interface
Product surfaces from the Synthex system.
Impact
From 30+ hours to ~3 hours per week.
The goal was to cut time spent on manual work and free advisors to focus on clients. The platform launched in the UK, expanded to South Africa, and started US expansion in early 2026, with Canada and Australia on the roadmap. The company secured a first round on the strength of the initial designs and a follow-on round for scalability and international expansion.
~90%
Reduction in time on manual tasks
Workflows
New capabilities as flows, not one-off features
UK → ZA → US
International expansion path
Prototype-led
Working React interfaces carried the first-round pitch — before production code