Thoa Tran Open to remote work · full-time or contract
Product Owner · AI and complex B2B

I turn messy product problems into specs, rubrics and shipped fixes.

I own product and AI quality for AffordX, a lending platform for mortgage brokers and financial advisers. I scope what gets built, write the tickets, test AI models before they reach clients, and trace support issues back to the product. Before that, I spent nine years designing B2B software for investment funds, supply chains, hospitals and telecom operators.

See selected work LinkedIn ↗
Model evaluation before a vendor upgrade 3 runs per model
Decision: pick a model per feature, not one model for everything. From an evaluation I ran in Oct 2026. Model names withheld in the public version.
93 daysdefault data window I defended, so clients share only what lenders need
~1 weekPhase 1 scope for a second lender, by reusing data we already had
3 runsper AI model in my tests, which caught one ticking compliance boxes with no evidence, so it was kept out of form filling
9+ yrsdesigning B2B products before moving into product ownership

DomainsFintech & lendingInvestment analyticsSupply chain & EDIHealthcare operationsDevice security & telecom

Selected work

Three problems I owned end to end

Three recent problems from two and a half years on the same lending platform: designing its flows, owning the AI that runs on them, and keeping them working when real banks break. Screens are rebuilt with sample data, and client and bank names are removed.

Lending · Platform scope · Jun – Sep 2026First lender live · second in final testing

From one lender to a lender-agnostic platform

Problem
Brokers could submit personal loans to one lender. A second lender was coming, with no UI of its own and an API-only integration.
My role
Led the project: scope, phases and milestones, requirements, design and QA, working with the partner lender's team.
Result
The first lender's submission flow is live. The second lender's Phase 1 reused the existing flow and a standard payload, and is in final testing before go-live.
Decisions I'd defend in an interview
  • Kept application status and lender status as separate fields, so one application can later go to several lenders.
  • Added a ticket to confirm the API contract before any UI work, so engineers didn't build against a payload that might change.
  • Limited Phase 1 to data we already collect. No new fields, documents sent as attachments. Speed over completeness, on purpose.
  • Dropped my first idea, a plan modal inside the applications list, in favour of one unified Loans module.
Read the case study
AI quality · Regulated advice · 2026Shipped

Making AI trustworthy where a wrong answer is a compliance risk

Problem
AI meeting summaries, client letters and form filling for financial advisers. In regulated advice, a fabricated detail isn't a typo. It's a compliance issue.
My role
Owned AI output quality: templates, prompts and bug tickets. Ran the model evaluation and recommended the setup.
Result
The model I recommended passed 20 of 21 answer key checks. The team shipped a per-feature setup before the vendor retired the old model.
Decisions I'd defend in an interview
  • Tested every model on repeated runs, not one demo. One model ticked 9 compliance declarations on every run.
  • Fixed meeting summaries that falsely flagged disclosures as given, then cut maintenance to one template version.
  • Separated a bank's generic conditions from a client's real action items, so approved clients weren't told to do things they didn't need to.
  • Required template-agnostic fixes for form filling, because it's sold as an API for any bank form.
  • Flagged that long financial documents were being skipped, which could treat self-employed income as zero, and proposed a visible warning.
Read the case study
Open banking · Incident & support · May 2026 – nowShipped

Keeping brokers' trust when open banking breaks

Problem
A large incident right after open banking went live, then a steady stream of bank-side issues that brokers saw as our bugs.
My role
Support triage, root-cause analysis, product fixes and the words brokers saw.
Result
Brokers kept informed through the launch incident until the team's fix landed, then a support playbook and plain-language guidance for recurring issues.
Decisions I'd defend in an interview
  • Designed a dismissible banner during the incident: open about the problem, without alarming brokers mid-application.
  • Defended the 93-day default statement window. It gives lenders enough history without exposing older client data. Longer isn't better.
  • Found why statements went missing: a bank's statement cycle has to fit inside the consented window. Turned that into guidance brokers could act on.
  • Wrote a support playbook: hypothesis, diagnostic question, evidence, no blame, workaround. Recurring issues go back to the product backlog.
Read the case study
Security QA

Permissions are product behaviour

I test IDOR and API-hardening regressions environment by environment, from Dev to Prod, before releases go out.

Release management

One tracker from Test to Prod

Every ticket in a deployment batch tracked through Test, Demo and Prod, so nobody has to ask what shipped.

Systemic bugs

One pattern, not four tickets

Spotted that a "missing fields" count drifted from field validation across four sections, and logged it as one systemic issue.

Earlier work

New domains, same method

Before fintech I worked on products for investment funds, telecom operators, warehouses and a hospital group. Each time I learned the domain the same way: map who does what, which data moves between them, and where it breaks.

Leonardo · Investment analytics · UK · 2019–23

Making a company's track record readable before a funding decision

Funds review a company's capability before they commit capital. I turned technical and business requirements into the analytics and data visualisation that review runs on.

Carries over: dense data behind an expensive decision. The same question as my AI evaluation work: what does the reviewer need to see to trust it?

Trustonic · Device security & telecom · UK · 2019–23

Getting operators' customers started on a platform on 350M+ devices

Onboarding and support channel for a device-security platform used by telecom operators in 50+ countries: video and audio guides plus technical documentation.

Carries over: support content that answers the question before it becomes a ticket, the same instinct behind my open banking playbook.

DiCentral · Supply chain & healthcare · US · 2017–19

B2B tools for warehouses, automotive suppliers and a hospital group

Web and mobile products at a Houston-based EDI company: a hospital group's supply tool, an automotive suite redesign, warehouse management, and role and contact management.

Carries over: many parties, many roles, one data flow. The groundwork for lender integrations and permission testing.

How I work

How I solve problems

Reproduce before I diagnose

I start from what the user saw and the evidence they can give. My tickets state observed facts; my guesses about root cause go in the comments.

Confirm the contract, then build

For integrations, the API payload is agreed before anyone designs a screen. It's the cheapest point to change your mind.

Judge AI on repeated runs

One good output proves nothing. I run each model several times per feature and score stability, fabrication and omission.

Async by default

Years of remote work with UK, US and New Zealand teams. Paste-ready tickets, short written updates, and a few hours of overlap every day.

Ho Chi Minh City · me--:--
London · daytime overlap--:--
New York · morning calls--:--
Design background

A product owner who can prototype her own specs

I spent nine years as a product designer before taking on product ownership. That means I can turn a spec into a clickable prototype myself, talk through edge cases with engineers in their terms, and check that what ships matches what we agreed.

  • Personal loan flow, designed from scratch through nine iterations2026
  • 15-screen HTML prototype to align product, engineering and QA2026
  • Analytics interface for investment funds, for a UK client2019–23
  • B2B product suite for a US supply-chain company2017–19
Financial plan workspace comparing a baseline and a second scenario over a client's lifetime
Financial plan workspaceLifetime projections and side-by-side scenarios, so advisers can show a client the cost of a decision.
AI form filler panel with three steps: select a form, loan structure, and reference sources
AI form fillerThree steps, with the sources the AI may use made explicit.
Client income and expenses view with mortgage and loan details
Client cash flowIncome, expenses and loans in one view, with summary cards above the detail.

Screens from the live product, cropped. Addresses and bank names blurred.

Experience

Where I've done it

2024 – now
Product Owner & Product DesignerPioneer Finance Group · AffordX · NZ, remote

Took a broker and adviser platform from one lender to a lender-agnostic flow, owned AI output quality, and ran support triage, security QA and releases.

2019 – 2023
Senior Product DesignerTP&P Technology, now AGEST

Owned UX for two UK clients, Leonardo and Trustonic: analytics for funds assessing companies before investing, and onboarding support for a device-security platform on 350M+ devices.

2022 – 2025
Professional Design CoachKeyframe Multimedia School

Taught product and UI design in Ho Chi Minh City.

2017 – 2019
UI & UX DesignerDiCentral, now TrueCommerce

Web and mobile tools for supply-chain, warehouse and hospital operations at a Houston-based EDI company serving 35+ countries.

2011 – 2017
Graphic → UI designerAl Fresco's Group, Vu Thao Computer

Print and brand work, then enterprise web and mobile layouts.

About

Hi, I'm Thoa.

I like the part of product work where something is broken and nobody knows why yet. I'm at my best between the customer, the code and the spec: finding the real cause, deciding what to fix first, and writing it down so the team can move.

I'm looking for a product owner or technical product role at a startup working on AI, fintech or complex B2B products such as logistics, health or data-heavy SaaS, ideally with teams in the US or UK. I'm based in Ho Chi Minh City and work remotely, with daily overlap for UK mornings and US calls.

Outside work: I founded Lotus Organization and represent the board of Nhà Cộng Đồng Thanh Liên, which gives free housing and meals to people in cancer treatment in Vietnam.

Ways to work together

  • Full-time, remoteProduct owner or technical product role
  • ContractProduct ownership for an AI, lending or complex B2B product, part-time or full-time
  • Freelance projectsAI output evaluation, clickable prototypes, UX audits

Skills

  • Product scoping
  • Requirements and user stories
  • Data dictionaries and field mapping
  • Integration requirements
  • AI output evaluation
  • Prompt and template design
  • UX and UI design
  • Interaction and motion design
  • Security QA
  • Release management
  • Support triage

Tools

  • Figma with AI plugins
  • FigJam
  • HTML prototypes
  • Jira and Confluence
  • Claude and Gemini