FlightHub · Solutions Lab · 2025
Dispute Defender
A chargeback system that gathered booking and payment evidence, applied reason-specific rules and drafted a response package for analyst approval.
Duncan Anderson · AI systems engineer · Montréal
AI systems that hold up inside real operations.
I design and build review-first AI for teams that run on email, spreadsheets, PDFs and exceptions. I spent four years inside FlightHub's fraud and AI teams building systems that moved real money. Now I build them for operations teams that can't afford a demo that breaks.
- 4 yearsbuilding fraud, dispute and pricing systems inside FlightHub
- 22 daysfrom first commit to live billing on The Lineup, a paid ML product I built and run
- EN / FRdiscovery, delivery and documentation in English or French
Start with a ten-day Readiness Sprint.
Before anyone builds, we map one costly workflow end to end, decide what AI should draft and what people must approve, and write the pilot brief. You keep every document whether or not we build together.
From CAD 6,500 · fixed before we start · 10 business days · 6 deliverables
How the sprint works →Systems in production.
Founder · 2025–
The Lineup
22 days
A web and iOS product that puts player projections, odds from 80+ bookmakers and recorded results in one place, with paying customers within 90 days of launch.
Trauma Therapy Group · Client · 2026
Trauma Therapy Group Publisher
Approved Google Docs become review-ready WordPress and Elementor drafts with SEO fields and imagery, and a leadership dashboard brings Jane reports and bank exports into one weekly view.
How I build.
The source stays attached.
Every draft links back to the records it came from, so a reviewer can check it in seconds.
Human authority is explicit.
The system drafts, sorts and assembles. Named people approve, and the system records who approved what.
Exceptions are designed, not discovered.
We list what goes wrong before building, and every exception has an owner and a path.
Start bounded.
One workflow, one team, one metric. Expand when the first one is trusted.
Measure operating value.
Hours returned, errors caught, money recovered. Not model scores.
Duncan Anderson
I'm an AI systems engineer in Montréal. I started in frontend development at CHU Sainte-Justine's research centre, moved into fraud analytics at FlightHub, and spent four years there learning how high-volume operations actually fail: at the handoffs, in the exceptions, and wherever someone has to rebuild context by hand.
In 2025 I moved into FlightHub's Solutions Lab to build AI systems for chargeback defence and deal ranking. Then I built The Lineup, a paid sports-analytics product, alone from first commit to live billing in 22 days. I still run it. Today I bring the same approach to operations teams: bounded scope, source-linked output, and people in charge of the decisions that matter.
- 2026–
Independent AI systems consultant
- 2025–
Founder & engineer, The Lineup
- 2025
AI Integration Specialist, Solutions Lab, FlightHub
- 2021–2025
Data Analyst, Loss Prevention, FlightHub
- 2019–2021
Frontend Developer, CHU Sainte-Justine Research Centre
Questions
What happens to our data?
It stays in systems you control wherever possible. I work with the least access the job needs, I don't train models on your data, and I document every place data moves before anything is built.
Which AI tools do you use?
Whatever fits your stack and risk profile: OpenAI, Anthropic and open models, plus the tools you already run. I'm independent and take no vendor referral fees.
Who owns what you build?
You do. Code, prompts, configuration and documentation are handed over at the end of every engagement.
What if the sprint says don't build?
Then you've spent ten days, not a quarter, finding out. You keep the workflow map, the exception list and the recommendation.
What happens after a build?
Every build ends with a handoff: a runbook, a named owner on your side, and a review of the first live weeks. Ongoing support is optional and quoted separately.
Do you work in French?
Yes. Discovery, delivery and documentation in English or French.
Have a workflow that eats your team's week?
Tell me about it in 30 minutes. If AI isn't the right fix, I'll say so.