Systems that are in use today.
A selection from over 200 projects. Each began with a process that took too much manual work and ended with a system built around exactly that process.
Companies and institutions we have built for
Clients in their own words
Selected projects
-
Construction controlling with automated invoice processing
1 system instead of inbox, spreadsheets and separate tools, across all entities and projectsStarting point
Subcontractor invoices arrived as email attachments and were entered by hand. Budget vs. actuals lived in spreadsheets that quickly went stale, with cash-flow planning kept separately.
What we built
Invoices are fetched automatically, read by two AI models and assigned to the right project and trade. A live dashboard shows budget, actuals and cash flow across all entities and projects.
Result
One system instead of inbox and spreadsheets. Variances per trade are visible at once, and the cash-flow forecast runs on real numbers.
-
Namaqal AI: speech recognition for Somali
−33 % processing time after the model update. Maay dialect from 0 to 9.4 out of 10Starting point
Somali audio had to be transcribed and translated into English by hand. There was no usable system for the Maay dialect.
What we built
A dedicated speech recognition system that turns Somali audio into text and translates it into English, with a review workflow for the team and a metrics dashboard.
Result
Live in production. After the latest model update the Maay dialect is usable for the first time, and processing time fell by about a third.
-
Sales configurator for solar systems, from quote to order confirmation
1 pass from system configuration to quote PDF with economic analysisStarting point
Sales worked with an older configurator, and the calculation lived in an Excel model. Quote, economic analysis and datasheets came from different sources.
What we built
A configurator for the sales team: pick modules, inverter, storage, mounting and add-ons, and the system is calculated. Out come a guide-price quote, self-consumption and economic analysis, datasheets and the order confirmation as PDF. The team maintains products, price categories and surcharges itself in the admin area.
Result
In use since 2025. The quote goes straight from the configurator to the customer by email. Built multi-tenant, so several companies work with it, each kept separate.
-
Market surveillance: spotting suspicious bidding in auctions
EU AI Act documentation included, with AI impact assessment and data quality reportStarting point
The platform runs auctions with many participants, and every bid triggers a chain of events. As a regulated marketplace it has to detect manipulation, and the existing surveillance relied on simple rules.
What we built
First an exploratory analysis of bidding behaviour. From it came an extended rule-based surveillance system with new indicators, for example for sharp quantity changes in bids or suspicious patterns across several auction rounds, plus experiments with ML approaches. Built and tested on red-teaming data, i.e. deliberately simulated manipulation attempts.
Result
The system is integrated into the client's AWS environment and handed over to the internal team, with the documentation the EU AI Act requires. Suspicious patterns surface through traceable indicators.
-
Comparing offers from 30 to 40 supplier emails a day
30–40 supplier offers a day, read and compared automaticallyStarting point
Dozens of offers arrived every day as PDF or Word, in Polish, Dutch or Spanish. Prices and cuts were compared by hand.
What we built
Emails are read automatically, prices and products recognised and normalised across languages. A dashboard lines up the offers, and a chat search finds past offers.
Result
No more manual sorting in the inbox. The team sees at a glance who has the best offer today.
-
From lab report to draft expert opinion
Seconds instead of hours to draft an expert reportStarting point
Lab reports arrived by email and were copied into the project system by hand before an expert could even start the assessment.
What we built
Middleware reads reports automatically, matches them with the project system and drafts the opinion. Approval follows the four-eyes principle. Plus an internal Private GPT with report templates.
Result
Drafting a report takes seconds instead of hours. Experts review and approve instead of typing.
-
Automation around a legacy system, without touching it
6–9 min of manual work per transaction removed, without touching the legacy systemStarting point
The core ran on a VB6 system with MariaDB. Documents for every transaction were created and filed by hand, six to nine minutes each.
What we built
An integration platform monitors database and file server, generates PDFs automatically and archives them. Optionally, a local AI sorts incoming emails and tickets.
Result
Minutes per transaction became near-instant. The old code stayed untouched.
-
A dictation app becomes a platform for many practices
1 → many practices on one platform instead of single installationsStarting point
The dictation app for medical findings ran as a single installation. Customer management, billing and subscriptions were missing.
What we built
After the desktop app came a platform with a central backend, an admin dashboard for customers and revenue, a practice dashboard, recording in the browser and on mobile, usage-based billing and automatic updates.
Result
The vendor can sell the software to any number of practices and manage subscriptions centrally.
-
Quotes straight from HubSpot, in their own design
20+ pages quote generated automatically from HubSpot, preview equals PDFStarting point
Quotes of 20+ pages were built by hand and looked a little different every time.
What we built
A quote engine in HubSpot where the preview matches the final PDF exactly, plus an ROI calculator that factors in client value and VAT for Germany, Switzerland and Austria.
Result
Consistent quotes in a fraction of the time, straight from the CRM.
What clients say
-
“With their extensive expertise, they were able to effectively relieve our internal team, so we could focus on our core business.”
Jacek Zawisza CEO, kamioni -
“Thanks to the collaboration with DeepMetis, we successfully implemented our digitization project within just five months.”
Pascal Bolsinger CEO, ADREX -
“I was impressed by their speed and dedication.”
Michael Raven CEO, Cognity Ventures
And from the people who use it every day.
From conversations with our clients' employees. Translated from German, names withheld.
- Usability
“I find it nicely clear. It's not overwhelming. It's somehow self-explanatory, too. … It has a bit of an Apple vibe … and really simple.”
- Relief
“It's really nice how much easier it already makes things.”
- First impression
“That's really great … I love it. A real winner.”
- Support
“I've never had such good support anywhere!”
Your process could be the next project.
Show us in a system check where things get stuck. Within a few days you'll see a clickable mockup.