Managed AI on retainer · Toronto · 2026

Roux.

One senior engineer in Toronto. I build the automations underneath your business — then host them, watch them, and keep making them better. Three are already built and live, from $149 a month.

§02 · A small B2B software team · Early revenue, roughly a dozen peopleA build like this typically ships in one to two weeks, with the first leads landing inside the first week.
Illustrative example — not a real client

Cold-outbound pipeline (an Apollo-style stack, rebuilt)

~100 ICP-ranked leads a week, no SDR headcount added

  • Roughly 100 ICP-ranked leads processed a week, each with a written, not merged-field, opening line
  • Built to lift reply rates by replacing template outreach with a real personalization hook
  • Sized to cover the outbound work a first SDR hire would otherwise be asked to do

What this builds

A Claude-driven cold-outbound pipeline that ingests Google Maps Places + Hunter + a CRM, ranks ICP fit per lead, writes a one-line personalization hook, and inserts ready-to-send drafts for a rep to review.

  • ICP scraper from Maps Places API → ~400 candidate companies/wk
  • Claude enrichment: decision-maker name, role, tenure, recent funding/hiring/launch signal
  • Hunter verifies email deliverability (≥95% confidence threshold)
  • Claude writes the personalized hook against a handful of seed example quotes per ICP
  • CRM upsert, assigned to a rep, nothing sends without review
STACKPythonClaude APIGoogle Maps PlacesHunter.ioHubSpot CRMSupabaseGitHub Actions cron

RETIRESReplaces the kind of stack Apollo, Lemlist, and Clay cover today

§03 · A small DTC apparel brand · Founder-led, one person running day-to-day opsTypically ships in about a week, with the first briefing landing within days.
Illustrative example — not a real client

Daily ops briefing (Shopify + Klaviyo + Meta + Stripe)

One Doc at 6am instead of a Sunday spent in four dashboards

  • Replaces a recurring block of weekend dashboard-checking with a single daily read
  • Built to surface a ROAS or spend anomaly the same day it happens, not at month-end
  • Retires the need for a paid analytics-dashboard subscription

What this builds

A daily ops briefing that pulls Shopify + Klaviyo + Meta Ads + Stripe at 05:55 ET, drafts a one-page Google Doc with the top anomalies and top actions, and has it in the inbox before 06:00.

  • Cron at 05:55 ET hits Shopify Admin API, Klaviyo API, Meta Marketing API, Stripe API
  • Claude reads deltas vs. a 7-day baseline, flags moves outside the normal range
  • Top actions ranked by likely $-impact (ROAS drops > inventory > churn signals)
  • One-page Doc generated, formatted, and emailed via Gmail API
  • Weekly: a longer write-up with a chart of the week and a 12-week trend
STACKPythonClaude APIShopify AdminKlaviyoMeta MarketingStripeGmail APIGoogle Docs API

RETIRESReplaces the kind of analytics-dashboard subscription tools like Triple Whale or Daasity sell

§04 · A small developer-tooling startup · Seed-stage, a handful of engineersTypically ships in a little over a week.
Illustrative example — not a real client

CI spend monitor + cron audit

Built to catch a CI billing spike before the invoice does

  • Flags a cost spike the same day instead of at the next invoice
  • Ranks every scheduled job by projected monthly cost so the expensive ones are visible
  • Sized to replace the part-time job of someone whose only task is watching the bill

What this builds

A GitHub Actions billing anomaly detector plus a cron audit tool that ranks every scheduled job by cost-per-invocation × frequency and flags the over-runs.

  • GitHub Actions billing API polled every 15 min
  • Rolling 7-day baseline per workflow, alert on deviations outside the normal range
  • Slack notification with workflow + run-id + projected hourly burn
  • Cron audit scans .github/workflows/ + crontab + Kubernetes CronJobs, sorts by predicted monthly cost
  • Daily report: the most expensive crons with downsizing recommendations
STACKPythonClaude APIGitHub RESTPagerDutySlackPostgresRender

RETIRESReplaces the kind of dedicated cost-monitoring tier something like Datadog sells separately

§05 · A small multi-location dental practice · A few locations, roughly two dozen staffThe most complex build in this set — PMS integration plus insurance edge cases usually push it toward three weeks.
Illustrative example — not a real client

Pre-appointment intake + multi-channel recall

Built to cut no-shows and put recall outreach on autopilot

  • Sized to lift recall coverage from a couple of due patients in five contacted, to nearly all of them within a day
  • Front-desk intake time modeled to drop from most of a day a week to under an hour
  • Built to meaningfully reduce no-shows and recover chair time that would otherwise sit empty

What this builds

A pre-appointment intake system that drafts a short chart summary 24 hours before each appointment, plus a multi-channel recall automation (SMS / email / voicemail) for patients who are due.

  • 22:00 nightly: pull tomorrow's schedule from a practice-management-system API
  • For each patient: chart history + notes → Claude drafts a short summary for the front desk
  • Insurance pre-check via an eligibility lookup
  • Recall pipeline: SMS → email → voicemail-drop, in sequence
  • Confirmation rebooking: SMS reply auto-books in the PMS
  • 06:00 office-manager dashboard: today's appointments, intake-delivery status, recall coverage
STACKPythonClaude APICurve Dental APITwilioPostmarkSlybroadcastTelus eHealthSupabase

RETIRESReplaces most of the front-desk hours spent on manual recall calls

From the studio

Built by Thomas. Toronto.

Roux is one person. I'm a senior engineer who spent the last decade shipping production software at SaaS companies you've used, and the last two years building Claude-assisted tooling for the kinds of automations on this page. I don't hand the work off to a junior, don't subcontract strategy, and don't run a content team. When you hire Roux, I'm the one who scopes it, builds it, and is on the hook when it breaks at 03:00.

Based in Toronto. Working with teams in the US and Canada. Engagements priced and invoiced in USD.

Thomas Peng · Toronto · May 2026

Pricing · USD

Two ways in. The audit is free either way.

Managed automations
$149
USD / month, per automation

Twenty jobs that are the same in every business — lead replies, inbox triage, quotes, chasing invoices. Three are already built and live; the rest are on the way, so you pay a fraction of a custom build.

  • Hosted, monitored, and tuned by me
  • Nothing sends without your approval
  • Three for $349/mo
  • Cancel any month
Browse the twenty →