meta
Vatsal Shah

Fractional AI Engineering

Embedded part-time AI engineering for teams that need to ship AI features but aren't ready to hire a full-time AI engineer.

Talk to Vatsal

A retainer-shaped engagement, not a project. We work inside your team — your Slack, your repo, your standups — for a defined number of days per week, shipping AI features alongside your existing engineers. You get an experienced AI engineer in the loop without the 6-month hiring cycle, and you keep the option to hire full-time later with a clean handoff path built in.

Fit for

  • Canadian companies with an in-house engineering team that needs AI capability, not full ownership of the AI roadmap
  • Teams shipping AI features regularly (monthly or faster) where the volume justifies recurring engineering time
  • Operators planning to hire a full-time AI engineer in 6–12 months and wanting a working system to hand them when they arrive

Not a fit for

  • One-off projects with a defined start and end — a fixed-scope engagement is cleaner and usually cheaper
  • Teams without any in-house engineering — we work alongside engineers, we don't replace them
  • Companies wanting a fully outsourced AI team — that's a different operating model

What you walk away with

  • A defined weekly commitment — typically 1, 2, or 3 days per week — with a named engineer (often Vatsal) embedded in your team
  • Active participation in your standups, code reviews, and planning, so we're shipping AI features as a member of your team rather than a vendor
  • Architecture, code, evals, and monitoring for the AI surfaces your team owns — model selection, prompt design, retrieval, integrations, production hardening
  • Knowledge transfer to your existing engineers built into the weekly cadence — pair-programming sessions, design docs, internal walkthroughs
  • Quarterly checkpoints to revisit scope, time commitment, and the path to full-time hiring
  • A handoff plan from the first week so the relationship can wind down cleanly when you bring on a full-time AI engineer

How the engagement runs

  1. 1

    Week 1 — Onboarding · Week 1

    Repo access, Slack access, intro to the team. Read the existing AI code, the roadmap, and the active issues. Pair with one of your engineers on a small task to calibrate the working style.

  2. 2

    Weekly cadence · Every week

    Defined days on, defined days off. We attend the standups for the days we're on, ship in the same repo your team ships in, and review PRs the way any other engineer would. No separate vendor process.

  3. 3

    Monthly review · Every month

    30-minute sync with your engineering lead — what shipped, what's blocked, where the priorities are shifting. Adjust the weekly focus before it drifts.

  4. 4

    Quarterly checkpoint · Every quarter

    Revisit scope, time commitment, and the hire-vs-extend decision. If a full-time hire is coming, we start preparing the handoff. If scope is growing, we discuss adding hours or a second engineer.

  5. 5

    Handoff · When you hire

    Two to four weeks of overlap with your new full-time AI engineer. Pair sessions, architecture walkthroughs, runbook handoff. We exit clean and your hire owns the system end to end.

By industry

For Canadian Financial Services

Embedded AI engineering for in-house teams at credit unions, MGAs, and fintechs — shipping copilots, document automation, and retrieval features alongside your existing platform engineers. We work inside your repo and your change-control process, not around it.

For Canadian Healthcare

Embedded AI engineering for product teams at healthtech companies and clinic networks — shipping documentation helpers, intake assistants, and EMR-integrated features alongside your in-house developers. Works inside your provincial framework and your existing release cadence.

For Canadian Customer Operations

Embedded AI engineering for SaaS and B2B product teams — shipping AI features inside your existing application alongside your engineering team. Useful when you have product-market fit and an AI roadmap but no AI engineer on staff yet.

Selected work

Recent fractional AI engineering work.

Read more in our engineering log →

Frequently asked

When does fractional beat hiring full-time?
When you need senior AI judgment now but the role doesn't yet justify a full-time salary — usually that's the first 6–12 months of building AI into the product. Once the AI surface area is large enough to keep someone fully busy, full-time is cheaper. We're explicit about that conversation at every quarterly checkpoint.
How many days per week?
Most engagements start at 1 or 2 days per week and adjust at the quarterly checkpoint. Some teams grow to 3 days during a launch window and step back to 1 day afterward. We don't do less than 1 day per week — anything smaller doesn't build enough context to be useful.
What's in scope and what isn't?
In scope: AI feature design and build, evals, monitoring, model selection, prompt design, retrieval, integrations, code review, pair programming with your team. Out of scope: managing your engineers, owning your roadmap, on-call rotation outside agreed hours, non-AI engineering work. We can flex the boundary, but it's worth being explicit upfront.
What does it cost?
Monthly retainer pegged to the days-per-week commitment. Pricing is shared on the first discovery call once we understand the scope and the cadence. Inference costs are billed directly to your accounts — we don't mark up tokens.
Who owns the IP?
You do. Standard work-for-hire contract — everything we write lives in your repository and belongs to you. We retain no rights to the code, the prompts, the evals, or the architecture.
What happens when we hire a full-time AI engineer?
We've designed the engagement around that path from day one. Two to four weeks of overlap, pair sessions, architecture walkthroughs, runbook handoff, then a clean exit. Some clients keep us on at a reduced commitment afterward for advisory work; some don't. Either is fine.
Where does the code run?
In your environment. We work inside your cloud, your repo, your secrets management. For Canadian data residency, we use AWS ca-central-1, Azure Canada Central, or Bedrock Canada. Inference providers chosen with your team based on cost, latency, and residency requirements.
Who does the work?
Typically Vatsal or one of two named engineers on our Toronto-based team. The engineer is consistent for the duration of the engagement — you're not getting rotated through a bench. The full team is named on our team page.