Five systems, described in full.
We show exactly what was built and how it works, in detail. We do not show before/after savings for anything that was not measured, so most of these briefs carry no hours-saved figure.
- Built to SOC 2 controls
- GDPR aligned · DPA on request
- Private AI on your own servers: yourtrained.ai
Invoice reconciliation, run on our own books
Code2b, our own back office
Supplier invoices arrived by email throughout the month and had to be matched by hand against the bank statement, one line at a time.
- In
- Supplier invoice PDFs collected from a folder
- Does
- Each invoice is hash-verified, renamed vendor-date-original, and filed into Invoices/year/month. The pipeline then matches each invoice line by line against the monthly bank statement.
- Out
- A reconciliation report with a matched table and an open table.
- Stack
- Claude API, n8n
Human checkpoint
Every unmatched charge is listed with a reason, never guessed.
Measured · August 2026
- 26 charges, 24 documented and matched, 2 open worth €18.10
- Caught: a full refund that needed a credit note before it could be expensed
- Caught: one invoice that covered three separate card charges
- Caught: a subscription billed in arrears that would have landed in the wrong month
- Caught: duplicate renders of one payment filed as several
Operations platform for an insurance brokerage
A national insurance brokerage, one of the largest teams in its market
Clients, deals, and the daily tasks that move a deal forward lived across spreadsheets and shared inboxes, across five departments, with no single source of truth.
- In
- Clients with a parent and child hierarchy, and every insurance deal
- Does
- A nine-stage deal pipeline, a task auto-created at every stage and routed to the least-loaded person in the responsible department, contracts driving client status automatically, an append-only audit trail enforced at the database, an SLA worker coloring every deal green, yellow, or red and escalating breaches to the manager, and role-aware dashboards.
- Out
- A running operations platform: pipeline board, client 360, contracts, and manager dashboards.
- Stack
- Go, Postgres, Next.js
Human checkpoint
The Client Decision stage (accept creates the contract, reject archives the deal) and manager escalation on any SLA breach.
Measured
- From the first workshop to a verified MVP in about a month.
- Not live yet: no operational numbers to report.
Anonymised at the client’s request.
Read the full briefContract intelligence for a regional operator
A regional operator, via a delivery partner
A high volume of contracts, including tenders, arrived as unstructured PDFs that had to be read and keyed by hand before anything downstream could use them.
- In
- Five document types including tenders, as Greek PDFs
- Does
- Five document types including tenders, extracted from Greek PDFs at roughly 90%+ field accuracy, with a mandatory human confirmation step before anything is saved, and counterparties deduplicated on tax number.
- Out
- Structured, confirmed records, saved only after a person signs off.
- Stack
- Claude API
Human checkpoint
The mandatory human confirmation step before anything is saved.
Measured
- five document types including tenders, at roughly 90%+ field accuracy.
Anonymised at the client’s request.
Rate intelligence engine for RoomPulse
RoomPulse
RoomPulse needed rate parity and pricing signals across a large hotel footprint, fed by data that most source platforms actively try to block from being collected.
- In
- Rate and availability data across 8 hotel platforms
- Does
- Collectors run on serverless workers and a queue, behind a proxy pool and an anti-bot service, feeding rate parity and pricing.
- Out
- A live feed serving 800+ hotel clients, with integrations into 2 hotel systems.
- Stack
- Serverless workers, a queue, a proxy pool
Human checkpoint
The pipeline runs unattended with monitoring and alerting; RoomPulse reviews parity and pricing signals downstream.
Measured
- 3 years running in production.
- 20M+ hotel rates processed daily, 150+ countries, RoomPulse engine.
Automations and data for Vistalink
Vistalink
Vistalink needed its booking automations and data pipelines built and kept running, so its own team could stay focused on the product.
- In
- Booking events and data feeds
- Does
- Booking automations, parsers, backend logic
- Out
- Bookings and data that stay in sync across Vistalink’s systems.
- Stack
- n8n, custom processors and parsers
Human checkpoint
Automation output feeds into Vistalink’s own review flow before it goes live.
Put your system on this page.
Book a 15-minute call. We scope the highest-impact build in your business, quote a fixed fee up front, and ship in 2 to 4 weeks.
- Built to SOC 2 controls
- GDPR aligned · DPA on request
- Private AI on your own servers: yourtrained.ai