Applied AI studio — Amsterdam
Meridian designs and ships AI assistants and automations that your team actually uses — grounded in your data, checked against your rules, and live in weeks, not quarters.
Live example An invoice automation, from inbox to approval. Sample data.
accounts@harbourfreight.example
Invoice HF-20931 — October haulage
Hello, Please find attached invoice HF-20931 for the haulage we completed in October under PO-7714. The total due is €4,860.00, payable by 28 November. Kind regards, Accounts team, Harbour Freight B.V.
Extracted · press a field to see its source
Checks
DraftBook €4,860.00 to 6120 Haulage · cost centre OPS-2 · pay on 28 Nov (not drafted yet)
Most AI projects stall after the demo. Ours are designed around one job your team does every day — so here is the screen people actually work in, not a diagram of a brain.
Chat and voice assistants grounded in your documents and systems. Every answer cites its source, and anything uncertain goes to a person.
Support assistant · internal preview
Can a customer return a sale item after 30 days?
Sale items can be returned within 14 days of delivery, unless they are faulty1.
Faulty items are covered by the two-year legal guarantee, so route those to a warranty claim instead2.
Automation handles the clean cases. Anything that fails a rule stops and waits for the right person — and every run leaves a trail you can audit.
6% over the PO. The amount is outside tolerance. The run stops, and Finance gets the invoice, the PO and the difference side by side.
Private deployments, no training on your content, and access that mirrors your existing permissions.
Every system ships with a test set and a quality score, so improvements are proven, not felt.
Assistants show their sources and ask for approval where it matters. Nothing is a black box.
One job, one team, one fixed scope. Each stage ends with something you can use or inspect.
Week 1
We sit with the team, map the work and pick one measurable job worth automating.
You getA map of the job, with today’s time per task
Weeks 2–3
A working version on your real data, tested against a scored set of examples.
You getA prototype and a 120-case test set
Weeks 4–5
Guardrails, permissions, logging and the edge cases the prototype skipped.
You getReview queue, audit log, access rules
Week 6
Rolled out to real users with training, a dashboard and a plan for the next job.
You getLive system and a shortlist for the next job
Three illustrative examples of the kind of job we take on, measured the same way before and after launch.