I build automation that takes the manual work off people's desks.

I'm Samuel, a Senior Automation Architect in Cheshire. I design AI agents, telemetry pipelines and small, sharp tools, and I've worked my way up from SQL support desks, so I build for the people who have to use the result.

A pipeline I run today: usage data from 50+ products goes in, early churn warnings come out.

Skills

AI and agentic automation
  • Building agents
  • Agentic RPA
  • Intelligent workflows
  • n8n
  • BrowserUse
  • Tavily
  • Workato
  • Power Platform
  • Embedding AI tools
  • API-driven integrations
Data and analytics
  • SQL
  • Supabase
  • Automated ingestion pipelines
  • Data lake infrastructure
  • Databricks
  • Power BI
  • ClicData
Monitoring and telemetry
  • Predictive usage infrastructure
  • New Relic
  • Coralogix
  • App Insights
Development and tools
  • Python for data automation
  • HTML
  • JavaScript
  • CSS
  • Git
  • Azure
  • Azure DevOps
  • Jira
  • Asana
Leadership and delivery
  • Stakeholder leadership to board level
  • Agile, Lean and Waterfall delivery
  • Incident, problem and change frameworks
  • Operational excellence
  • FranklinCovey leadership

Personal projects

Two tools I designed and built end to end in my own time. Each is a hand-written front end in a single HTML file, with n8n workflows doing the AI and data work behind it.

OEMate

An AI parts assistant for the motorcycle workshop counter

Enter the bike on the ramp and ask for a part. An n8n AI agent returns the OEM number, whether the fitment is verified, any supersession and equivalents, and the cheapest listings including postage.

  • n8n AI agent
  • Webhooks
  • Vanilla JavaScript
  • Structured LLM output

JobScentR

AI job matching for job seekers

Upload your CV and it collates your skills and roles for you, compares them against listings from all the popular job boards, and returns ranked matches with a reason for each one.

  • n8n
  • Supabase
  • Gemini
  • CV parsing
  • Accessible wizard

Experience

  1. Oct 2024 – Present

    Senior Automation Architect

    The Access Group
    • Enterprise telemetry infrastructure. Programme managed a business-critical initiative automating data capture from 50+ systems into a centralised data lake, creating a universal adoption scoring model across disparate SaaS products.
    • Customer retention and account health. Designed and deployed predictive health-degradation alerts and embedded them with Customer Success teams, giving measurable GRR improvements and higher tNPS.
    • Intelligent agentic workflows. Architected automated sentiment tracking inside core products, removing manual CS admin while surfacing real-time satisfaction metrics to mitigate churn.
    • Executive alignment. Championed operational metrics and project health transparency at board and director level, aligning previously siloed management teams.
  2. Jun 2023 – Oct 2024

    Operations Analyst

    The Access Group
    • Automated onboarding programmes. Architected in-product content delivery that streamlined client onboarding, cut Professional Services operating costs and raised onboarding utilisation.
    • API and system integration. Built unified integrations between Salesforce CRM, Qualtrics XM and core product APIs, automating data syncs and removing manual admin.
  3. Jan 2022 – Jun 2023

    Onboarding Consultant / Frontend Developer

    The Access Group
    • HTML and JavaScript. Built customer recruitment portal sites on the Volcanic platform.
    • Customer engagement. Worked directly with enterprise clients to translate business processes into scalable product configurations.
  4. Jun 2019 – Jan 2021

    Second Line Support Analyst

    The Access Group
    • Established a second-line helpdesk. Hired to set up and run second-line support for private and state education software.
    • Escalation and incident management. Led technical investigations with advanced SQL analysis to isolate and resolve critical application bugs.
  5. Sep 2015 – Jun 2019

    Support Analyst

    Touchretail (Swan Retail)
    • Database administration. Managed and optimised production SQL databases: data integrity, automated backups, security patching and proactive health monitoring.
    • Cross-functional support. Partnered with engineering and infrastructure to resolve core application errors and streamline escalation paths.

Get in touch

Have a process that eats your team's time, or a question about one of my projects? Send me a message and I'll reply by email.

mail@samuel-sanders.com

OEMate

Finding the right part for an older motorcycle means cross-checking fiches, superseded numbers and a handful of sellers. OEMate puts that behind one question at the workshop counter.

How it works

  • Bike + part
  • →
  • n8n webhook
  • →
  • AI agent researches
  • →
  • Structured JSON
  • →
  • Part cards + parts list

What it does

  • Bike-first context. The bike on the ramp is set once and sent with every question, with one-tap shortcuts for service kits, brake pads, filters, plugs and chain and sprockets.
  • Part cards built for ordering. Each part shows its OEM number in a large monospace face, its position, a clear Verified fit or Check before ordering badge, supersession and equivalents.
  • Cheapest first. Listings are sorted by price plus postage and trimmed to the best six.
  • A parts list that stays put. Add parts, adjust quantities and copy the list or export it as CSV. It survives a page reload.

Engineering decisions

  • Tolerant of LLM output. The response parser accepts plain text, JSON, arrays or JSON wrapped in code fences, so a slightly off-format agent reply still renders.
  • Made for a workshop LAN. Copy falls back to a hidden textarea where the secure clipboard API isn't available over plain http.
  • Safe by default. All agent text is escaped and only http(s) links are rendered.
  • Usable with gloves on. 44px touch targets, a slide-over list on phones, automatic light and dark themes, and plain-English errors that say where to look in n8n.
  • No framework, no build. One HTML file, under 200 lines, that runs anywhere.

JobScentR

Job hunting means retyping your skills into every site and trawling several boards for the few roles that fit. JobScentR reads your skills straight from your CV, searches the popular job boards in one go, and ranks every listing by how well it matches you.

How it works

  • Upload CV
  • →
  • Skills and roles collated
  • →
  • Popular job boards searched
  • →
  • AI scores each listing
  • →
  • Ranked matches

What it does

  • Skills straight from your CV. Upload a PDF or Word CV and Gemini, via n8n, collates your roles, skills and years of experience. No forms to fill in from scratch: you just confirm the roles, trim the skills and set location, work style, level and salary.
  • Every popular job board at once. One search covers listings from the popular job boards, so you aren't repeating the same search site by site.
  • Ranked matches with reasons. Each listing is scored from 0 to 100 against your profile and ranked best first, with a short "why it fits", search, score filters and pagination.
  • Keep more than one profile. Switch between saved profiles, for example one per type of role you're targeting, and edit or delete them at any time.
  • Cheap to run. Saved matches load instantly; the slower, token-costing AI search only runs when asked, or automatically for a brand-new profile.

Engineering decisions

  • No stale screens. Every request carries a sequence number, so a slow reply that lands after the user has moved on is ignored.
  • Built for real LLM replies. A recursive key search finds skills and roles wherever the model put them, including inside JSON strings.
  • Errors that tell you what to fix. Timeouts, CORS failures, empty bodies and wrong response nodes each get a specific message pointing at the n8n step to check.
  • Configurable back end. The n8n address is set in the app and remembered per browser, so it runs against any n8n instance.
  • Accessible throughout. Focus moves to each step heading, the step rail is keyboard navigable, status is announced to screen readers, and motion respects reduced-motion settings.