Your AI partner for growth and efficiency
Find the work your team dreads. Hand it to AI you can audit.
I sell outcomes, not tools. Every project is tied to one number we agree on up front: more customers, more value from each customer, or lower costs. Then I build the AI system that moves it, with tests, source citations and a human checkpoint where it matters.
Fundy Labs is founded by Hamish Anderson and based in Halifax, Nova Scotia. Working remotely with teams across the United States and Canada.
Three outcomes
Every project moves one of three numbers.
If a build doesn't clearly win more customers, raise what each customer is worth, or cut costs, I won't recommend it. Pick the complaint that sounds most like yours.
Approach
Pain first, technology second.
You get something useful at the end of every stage, and you can stop after any of them.
Find the pain
Short interviews and process walk-throughs with the people doing the work. We look for repetitive reading, copying, checking and chasing.
You get: a pain-point mapPick the winners
Each opportunity is scored on hours saved, risk, data readiness and effort. Some problems need a spreadsheet, not AI, and I will say so.
You get: a ranked roadmap with ROIProve it on your data
A working pilot on your real documents and systems, measured against an evaluation set your team agrees is the bar to beat.
You get: a pilot and its scorecardRun it for real
Production deployment with monitoring, regression tests and cost tracking, then handover and training so your team owns it.
You get: a system in daily useThe four blanks
No project starts until all four are filled in.
We walk through your operations front to back and find where friction first appears, which is the real constraint, not just the loudest annoyance. If we can't fill in these blanks, we don't have a project yet. Sometimes the honest answer is a plain workflow, because more AI means more cost and more risk.
This project is in the cut costs bucket.
The KPI is analyst time per memo.
The baseline today is about 4 days.
After 60 days we expect under 1 day.
Actual result: about 2 hours.
Services
What I build.
I work across the stack, from model choice and prompts to cloud infrastructure and the screen your team uses. I am not tied to any vendor; models are chosen per task on measured results.
Document AI
Pull facts out of contracts, reports, invoices and forms into structured data, with each value traced to the exact quote it came from.
Knowledge search (RAG)
Ask questions across your policies, past deals or technical docs and get answers with citations, not guesses.
Agents and multi-agent workflows
Assistants that look things up, use your tools and hand off to a person when confidence is low. Built for a defined job, not a demo.
Evals and guardrails
Test suites that tell you if a change to a prompt, model or data made things better or worse, plus tracing on every call.
AI-assisted engineering
Get your developers shipping faster with Claude Code and shared agent skills, with review gates matched to risk.
Fractional AI lead
Part-time technical leadership: strategy, vendor and model choices, hiring, and keeping AI projects honest about results.
Selected work
Results, written as baseline and after.
Every case study names its outcome, its KPI and where the number started, so you can judge it for yourself.
Investment memos from feasibility studies
Project finance platform · 20 enterprise customers
A multi-step pipeline reads 100+ page studies, extracts every fact with its source quote, verifies each one in three stages and drafts the memo. Analysts see a confidence score per fact, so they know exactly what to check.
- Outcome
- Cut costs
- KPI
- Analyst time per memo
- Baseline
- ~4 days
- After
- ~2 hours
Choosing the right model by evidence
Same platform · model evaluation
Ran a side-by-side evaluation of a system-one model (jev.ai) against the existing pipeline. It lost on extraction (78% vs 93% coverage), so it stayed out of that step, but it won on fact scoring and shrank what each generation call needs to read.
- Outcome
- Cut costs
- KPI
- Context per generation call
- Baseline
- 100%
- After
- 50 to 80% less
Faster delivery for a growing team
Consulting firm · engineering lead
Introduced delivery metrics and the practices behind them while growing the engineering team as the company scaled to about 40 people.
- Outcome
- Cut costs
- KPI
- Lead time for new features
- Baseline
- 2 weeks
- After
- 3 days
Proof I use it
Systems I run every day.
I build with the same AI tools I recommend. These are working systems, not slides.
Claude Code on a define, plan, build, verify loop
Shared instructions and a skills library that a team of five senior engineers builds on daily, with tests written first.
Evals that run on every change
Extraction recall, verification accuracy and hallucination audits run in CI whenever a prompt, model or data source changes.
Risk-tiered pull request gating
Low-risk changes that pass automated review ship without waiting. Anything touching data or APIs gets a human.
A guidance assistant that knows the next step
A tool-using agent that looks up where a user's project stands and retrieves best practice and prior deals to suggest what to do next.
How I build
AI your auditors, lawyers and customers can live with.
Every answer shows its source
Outputs link to the document and sentence they came from. If the system cannot find support, it says so instead of inventing an answer.
Measured before it is trusted
We agree what "good" means up front and test against it. The same tests run every time a prompt, model or data source changes.
People stay in charge
Confidence scores show your team where to rely on the draft and where to check. Review steps go where the risk is.
Your data stays yours
Data residency, private networking and customer-managed encryption keys when you need them. I have built under EU data-sovereignty rules.
About the founder
Hi, I'm Hamish.
I'm a senior AI engineer and former founding engineer who has spent two decades turning messy business problems into working software.
Most recently I built an AI project finance platform from zero to production, where it now supports €25B in deal flow. Before that I led engineering at a consulting firm, built payment APIs at a crypto lender, and worked on systems for Fidelity Investments and Commonwealth Bank of Australia. I studied mathematics and hold a graduate certificate in applied finance, so I am comfortable in the numbers as well as the code.
I started Fundy Labs to bring that same production-grade AI to businesses that don't have an AI team of their own. For larger builds I bring in a small bench of experienced engineers I have worked with for years. You keep one point of contact and one person accountable.
Ways to start
Four rungs. Each one earns the next.
Start with an hour and climb only when the last step paid off. Projects are priced against what the manual work costs you today, and I'll show you the math.
-
Rung 0 · 1 hour
Working session
$250 per hour
Hands-on time to teach your team a tool or set up a first automation. A low-risk way to see where AI fits.
-
Rung 1 · about 2 weeks
Audit
$1,500 to $5,000
We map your workflows, find the real constraint and rank what's worth automating. You get a written report and a proposal for the first project with the four blanks filled in.
-
Rung 2 · about 4 to 6 weeks
Project
Scoped after the audit
One focused build that proves ROI on your real data, measured against the baseline we agreed on.
-
Rung 3 · ongoing
Retainer
Scoped after the audit
Run, monitor and improve what we built, add the next project, and act as your part-time AI lead.
Prices in US dollars. Projects and retainers are priced against the cost of the manual work they replace. Example of the math: if a workflow takes 15 hours a week and we cut it to 1, that's about 700 hours a year back.
Questions
What people usually ask.
Do we need clean data or a data team first?
No. Most of the value I find sits in documents, emails and spreadsheets as they are today. Part of discovery is checking whether the data is good enough, and saying so plainly if it is not.
Which AI models do you use?
Whatever performs best on your task at a cost that makes sense. I have shipped with Claude on AWS Bedrock, open-weight models and specialist models such as jev.ai, and I choose between them with evaluations rather than preference.
Is it safe to put our documents into AI?
It can be, when it is set up properly. I can keep processing inside your cloud account and region, with private networking and your own encryption keys, so data is not used to train anyone's model.
We're a small business. Is this for us?
Yes. Some of the best payback comes from small teams where one person loses a day a week to admin. Start with a one-hour working session or an audit, which is sized to pay for itself if we find one good fix.
How do you set prices?
Against the cost of the work being replaced. If a task eats 15 hours a week and the system brings it to 1, we can both see what that is worth, and the project price will be a fraction of it. I'll walk you through exactly how I got to the number.
What if AI isn't the right answer?
Then I will tell you. Sometimes the fix is a better form, an automation or a process change. You pay for finding the problem, not for selling you a model.
Start with a conversation
Tell me what's slowing your team down.
A free 30-minute call. Bring one process that eats too much time. You'll leave with an honest view of whether AI can fix it, which of the three outcomes it moves, and roughly what it would take.