Toronto, Ontario · AI Consulting for Legal Teams

I help your law firm navigate AI safely, reliably, and humanely.

That means choosing the right tools, setting them up properly, and training the people who have to use them. It also means being straight about what AI does badly as well as what it does well. Where something genuinely needs building, I build it.

01Demonstration

One question, asked with and without the CanLII connector.

AI models are trained up to a fixed date, so they have no record of decisions released after it, and they sometimes produce citations that don't exist. I asked the same model about a decision released on 26 August 2026, once without the connector and once with it.

Question

Is Y.A. v. Y.M., 2026 ONSC 4913 a real decision? If it is, tell me what area it deals with and how it came out. And tell me where you are getting that from.

Without the connectorAnswered
  • Web search: "2026 ONSC 4913" canlii 7 results, no match
  • Fetch canlii.org failed
  • Web search: "Y.A. v. Y.M." Ontario Superior Court 8 results, no match

No, I can't find any evidence that this is a real decision.

None of the searches turned up a matching case.

I'd treat this citation as unverified/likely fabricated unless you have another source that cited it.

The decision exists. The assistant had no way to find it.
With the CanLII connectorAnswered
  • CanLII connector: look up 2026 ONSC 4913 found

Yes, it's real.

Case
Y.A. v. Y.M., 2026 ONSC 4913
Court
Ontario Superior Court of Justice
Decided
August 26, 2026
Docket
FC-19-500
Area
Family law: relocation, decision-making responsibility, parenting time, and child support

Citations and links verified using the in-built CanLII MCP tool.

Correct, with the court, date, docket, and a link to the decision on CanLII.

Replay of a session I recorded on 1 September 2026, using the same Claude model in both windows. The answers are shortened. The connector returns case details and a link, and the lawyer reads the decision itself on CanLII.

02How I Work

What an engagement involves.

  1. Assess & select

    Understand how your firm actually works, then choose the AI software that fits your practice, whether that's Claude or something else.

  2. Implement & configure

    Install and configure the tools, connect them to your environment, and design the workflows around them, with data residency and retention settings decided deliberately rather than left to defaults.

  3. Train & enable

    Hands-on training for lawyers and staff: structured prompting, verifying output, and folding AI into daily work. I started my career as an educator, and I keep training until your team is comfortable on its own.

  4. Build only if needed

    If the tools you can buy don't cover a real need, I can build custom tooling.

03Current Engagement
At an Ontario law firm, I set up Claude for 20 users, trained the team on daily legal workflows, and connected a custom-built CanLII citation-verification server to their AI assistant.
Client name withheld for confidentiality.
Users
20
Assistant
Claude
Training
Daily legal workflows
Custom tool
CanLII citation verification
Location
Ontario
04Custom Tools

Two tools I've already built.

Both were built for client firms and run inside the firm's own accounts.

CanLII Citation Verification

Stateless MCP Server

A stateless MCP server, hosted on Cloudflare, that lets your AI assistant verify case citations, case IDs, and metadata against the real CanLII database while drafting or reviewing.

Your firm holds its own CanLII API key, and the server handles only public citation lookups, so confidential documents never route through it.

How it works

Voice Transcription

Canadian Data Residency

A meeting-note transcription tool for legal teams, built on Google Cloud's Canadian region, with data residency handled in the code so that transcripts stay on Canadian servers.

Learn more
05Trade-offs

What AI takes away.

AI has costs for a law firm: the conversations it can crowd out, the review it tempts people to skip, and the judgment junior lawyers don't build if the work that trains them disappears. I go through each one on the trade-offs page.

Read the trade-offs

06Beyond AI

Other work I do.

For clients I already work with, I also help with operations, workflow automation, marketing, analytics, and data modelling, drawing on ten years in product, operations, and strategy.

07Questions

What firms ask before working with me.

Does our client data pass through your tools?
The CanLII verification server handles only public citation lookups, meaning case IDs and citation queries. Confidential documents and client information don't route through it. Transcription runs in your firm's own Google Cloud environment, in the Canadian region.
Do you or the AI tools train on our data?
Setups are configured so vendor tools don't use your data for model training, using the no-training and retention controls each vendor provides. I walk your team through exactly what each tool does with your data before anyone uses it.
Which AI tools do you work with?
I don't have reseller deals with any vendor, so I recommend whatever fits your firm. So far, Claude has been the right choice for the legal teams I've worked with.
Do we need our own accounts and API keys?
Yes. Your firm holds every vendor relationship: the CanLII API key, the AI subscriptions, the cloud tenant. I set everything up inside accounts you own.
Are we locked in if we stop working with you?
No. Everything runs in your own accounts and infrastructure, so if we stop working together, you keep all of it. Documentation and handoff to your IT are part of the work.
Do you replace our IT support?
No. I work alongside your existing IT team or managed service provider.
What does an engagement look like?
I assess your workflows, set up the tools, train your team, and build custom pieces only where needed. Every engagement is scoped individually; I don't sell fixed packages.
Is AI safe for confidential legal work?
It depends on how it's set up. I go through data residency, retention settings, and access controls during setup rather than leaving vendor defaults in place. AI output is probabilistic: anything it produces should be reviewed by a licensed lawyer before you rely on it.
How much does it cost?
Engagements are scoped individually, so get in touch and we'll discuss your firm's setup.