Skip to content
NDNicolas Desjardins, back to overview
Resume

AI automation · Greater Montreal, Quebec and Canada

AI automation for Canadian businesses

I'm Nicolas Desjardins, and Webman (webman.dev) is how I take on AI automation and full-stack work. I build AI agents, document processing and private knowledge-base search that connect to the tools your team already uses, so people stop retyping data, stop digging through manuals, and keep the final say on anything risky.

Get in touchEmail me

What AI automation services does Webman provide in Canada?

Five kinds of work, each with proof you can check. Measured results come from dated public records; the track record comes from my work at New Roots Herbal Inc.

  • Measured

    Document processing and data entry automation

    Invoices, purchase orders, forms and incoming emails become clean records in your accounting system, ERP or database, so nobody retypes them. Every total is re-added to the cent, and nothing comes out until it passes a validated schema.

    Typical uses

    • Supplier invoices and purchase orders
    • Intake forms and applications
    • Lead and support email triage
    • Receipts and expense reports

    Proof

    Invoice reader: 21 of 21 test invoices fully correct on the 7 October 2026 benchmark.

    See the proof: Document processing and data entry automation
  • Measured

    Private knowledge base AI and internal copilots (RAG)

    Staff ask questions in plain language and get answers from your own procedures, manuals and regulatory documents, with the exact passage and page quoted. When the documents do not cover a question, the assistant says so instead of guessing.

    Typical uses

    • SOPs and quality manuals
    • Regulatory and compliance documents
    • Product dossiers and technical specifications
    • HR policies and onboarding guides

    Proof

    Ask your manuals: 6 of 6 recorded questions answered with verified quotes, and the one question the documents did not cover was declined (14 October 2026).

    See the proof: Private knowledge base AI and internal copilots (RAG)
  • Measured

    Custom AI agents with human approval

    Multi-step back-office work, such as pulling data, drafting content and checking it against your rules, runs on its own up to the risky step. Then it waits for a person, and every decision is on record.

    Typical uses

    • Product listings for each sales channel
    • Compliance checks before anything is published
    • Order and inventory follow-ups
    • Drafted replies for a person to approve

    Proof

    Supervised operations agents: 6 scenarios recorded on 15 October 2026, every dispatch behind a recorded human decision, and a swapped probiotic dosage caught by a deterministic check.

    See the proof: Custom AI agents with human approval
  • Measured

    MCP servers and LLM integration

    Your systems become tools that AI assistants such as Claude, ChatGPT or Cursor can use safely through the Model Context Protocol (MCP): scoped access tokens, typed actions, and a confirmation before anything changes.

    Typical uses

    • Safe AI access to your CMS, CRM or database
    • Connecting models to internal APIs
    • Routing between AI providers with automatic fallback
    • API keys encrypted at rest

    Proof

    NextBlock CMS (Cortex AI): an MCP server with 56 tools, listed in the official MCP Registry as dev.nextblock/nextblock.

    See the proof: MCP servers and LLM integration
  • Track record

    Full-stack Next.js, Supabase and e-commerce development

    Websites, online stores and internal apps built to stay up: server migrations, headless CMS, custom API integrations, and the full-stack work an AI project needs around it.

    Typical uses

    • Next.js, React and TypeScript applications
    • Supabase and PostgreSQL back ends
    • E-commerce platforms and migrations
    • Laravel and PHP platforms

    Proof

    30+ websites migrated with zero downtime and hosting costs cut by over 40%, at New Roots Herbal Inc.

    See the proof: Full-stack Next.js, Supabase and e-commerce development

How do custom AI agents and MCP integrations connect to your existing tools?

Through the doors your software already has. Most business tools expose an API, a database, an email inbox or a file export, and that is all an agent needs.

  1. 01

    Pick one workflow

    Start with the task that eats the most hours each week, and write down its inputs, its rules, and the steps a person must approve.

  2. 02

    Turn each action into a typed tool

    Every action the AI may take, such as looking up an order or drafting an update, becomes a tool with a strict schema. Read-only tools run freely; tools that change data ask for confirmation first.

  3. 03

    Give it only the access it needs

    The assistant connects through an MCP server or a direct API integration with a scoped token, so it reaches the records it needs and nothing else. Keys are encrypted at rest.

  4. 04

    Keep a person on the risky steps

    The workflow pauses before anything is sent, published or changed, and records who approved it. Every run can be traced afterwards.

  5. 05

    Measure it on your own cases

    Before rollout, the system runs against a set of your real documents or requests and you see the results, the same way every project on this site is measured.

AI automation in Montreal and Quebec, across Canada, and remote worldwide

I'm based in the Greater Montreal Area and work in English and French. Companies in Montreal, Salaberry-de-Valleyfield and elsewhere in Quebec can meet in person; teams across Canada, the US and the rest of the world work with me fully remotely. Full-time roles, contracts and agency placements are all welcome.

Questions about AI automation

What is Webman?

Webman (webman.dev) is the name Nicolas Desjardins works under on the web. He is an applied AI developer and senior full-stack developer in the Greater Montreal Area, and the creator of NextBlock CMS and its Cortex AI MCP server. He first used the name for Webman Solutions, his web studio from 2016 to 2018.

Can AI connect to the software we already use?

Usually, yes. If a system has an API, a database, an email inbox or a file export, an agent can read from it and, with your approval, write to it. When a tool has no API, a scheduled export or a shared folder is often enough to start.

How do you keep the AI from making things up?

By design: answers quote the passage and page they come from, questions the documents do not cover are declined, extracted data must pass a validated schema before it is saved, and risky actions wait for a person. The projects on this site publish their measured results with dates.

Where does our data go?

That is decided with you before anything is built: a commercial model API, a provider hosted in a region you choose, or an open model running on your own server. The search index can live in your own PostgreSQL database.

Do you work outside Montreal, or outside Canada?

Yes. I work on site or hybrid in Montreal and the surrounding region, including Salaberry-de-Valleyfield, and fully remote anywhere in Canada, the US or elsewhere in the world, in English or French.

Can you join our team instead of working on contract?

Yes. I'm open to full-time roles, contracts, and placements through an agency.

How fast do you reply?

Within one business day. Describe the problem in your own words; you do not need to know the technical name.