Expertise
Our areas of expertise, in detail
Seven domains, and for each one what actually gets delivered. None of it is sold separately: these are the skills that go into a project, in whatever proportion it calls for.
Getting your shop cited by LLMs
An assistant recommending a product does not click a blue link: it reads a page, keeps a few facts and cites them. Rank is no longer the point — being the source that gets quoted is.
- Product, Offer, price and availability as structured data, readable without running JavaScript.
- An /llms.txt and plain-text pages: what an agent crawler reads, not what a browser does.
- Short, attributable answers — one question, one paragraph, one checkable figure.
- robots.txt set per agent: GPTBot, ClaudeBot, Google-Extended, PerplexityBot.
- Citation tracking: which questions bring your name up, and in which assistant.
Wiring your sales into Claude, OpenAI and Gemini
A catalogue exposed as a tool becomes a sales channel: the assistant searches, compares and prepares the order without leaving the conversation.
- A bespoke MCP server: catalogue, stock, cart and order status exposed as tools.
- Function calling on OpenAI and Gemini off the same foundation — one contract, three clients.
- Guardrails: tool scope, quotas, a call log, and a human sign-off before payment.
- Connectors for Shopify, WooCommerce, Stripe, or your own APIs.
- A sales agent on your own site: it answers, qualifies, and hands over to a person at the right moment.
Moving classic search rankings up
SEO has not gone away: it still pays for most of the traffic. And it turns on measurable things — what loads, what reads, what gets linked.
- Technical audit: indexing, canonicals, hreflang, sitemap, structured data.
- Core Web Vitals measured in the field rather than in a lab — LCP, INP, CLS.
- Metadata rendered server-side: without it a SPA shows a crawler nothing.
- Content architecture: one page, one intent, and internal links that back it up.
- Rankings and entry pages tracked month by month, with a list of what moved.
Automating what repeats
A task done by hand twice a week is a script that has not been written yet. Measure it first, remove it second.
- Quotes, invoices, chasers and exports triggered by an event rather than by a reminder.
- Tools kept in sync — CRM, accounting, spreadsheet, inbox — with no copy-paste.
- Document handling: extraction, filing, consistency checks.
- Scheduled jobs that are watched: a failed run speaks up rather than going quiet.
- Every automation stays readable and reversible — versioned code, not a black box.
Putting AI into the everyday
Useful AI is not a demo, it is something you open on a Monday morning. The question is always the same: which gesture does it remove.
- Search across your own documents, with the sources sitting next to the answer.
- Drafting in your own voice: product copy, standard replies, write-ups.
- Email and tickets sorted and pre-answered, with a person keeping the last word.
- Summaries of a meeting, a contract or a report, backed by the exact quote.
- A model picked per task, and the cost per call shown rather than absorbed.
Taking analytics back
A measurement that changes no decision is a cost. Install the minimum that answers questions asked in advance.
- Measurement with no third-party cookie, hosted in Europe, GDPR-compliant.
- Events named after your decisions: the question first, the instrumentation second.
- A dashboard readable in thirty seconds — entries, conversions, drop-off.
- Honest attribution: what can be claimed, and what cannot.
- A raw export available at any time: the data stays yours.
Designing a database that holds
A schema decided in ten minutes is paid for over years. Everything else — speed, reliability, backups — follows from it.
- Relational modelling: PostgreSQL by default, SQLite where it is enough.
- Versioned, reversible migrations, run in CI before they run in production.
- Indexes placed on real, measured queries rather than assumed ones.
- Encrypted backups, tested by restoring them — a backup never restored does not exist.
- Vector search where it genuinely earns its place (pgvector), and nothing beyond that.
All pages
archambault.a.alexis@gmail.com
https://studio.archambault.site/