Berlin Law Firm
Berlin · AI on their own hardware — It started as a question about scanning the mail. Through the consulting it became a system: an AI pipeline that reads, understands and files every document — and never lets a deadline slip. In build now, entirely on the firm’s own servers.
— The project
One small ask. A system — and a partnership.
Context
For a law firm, a missed deadline isn’t an inconvenience — it can be malpractice. And a German law firm lives on paper: every day the post brings court letters, briefs and notices, each scanned by hand, named by hand, filed by hand into the right case folder. At a small Berlin firm — under ten people — that work fell on people with better things to do, and every manual step was a place where a date could hide.
We met through an academic introduction. While the engagement is live, the firm stays unnamed at its own request — the discretion any law firm expects, and the same we extend to every client.
The ask
The first question was modest: can the scanning and filing be automated? Over a series of consulting sessions we took apart how mail actually moves through the firm, what a brief contains, and where the hours go. That diagnosis changed the scope: if a machine was going to read the documents anyway, it could do far more than sort them.
What we’re building
Read
A document-specialist AI-OCR engine — not a generic scanner — turns every page into clean, structured text. Which means: no retyping, and no bad scan swallowing a fact.
Understand
A language model reads each document and extracts what the firm actually needs — case number, parties, deadlines, a working summary — as structured data, not loose notes. Which means: a brief arrives and the case file already knows what’s inside.
Verify
Before anything is saved, a second model checks the first one’s work against the firm’s standards — the four-eyes principle, in software. Deadlines get special treatment: a brief can carry several dates, so dedicated validation steps decide which one is binding — and where the law requires it, the system calculates the operative deadline instead of trusting what’s printed.
File and log
The document lands in the right case folder automatically, and every step is written to an audit log the team can inspect. Which means: the firm keeps what automation usually takes away — the ability to check.
On their own servers, in Germany
Legal documents are as sensitive as data gets, so the architecture question came before any build: cloud or local? We consulted on both, openly. The decision: everything runs locally, on hardware in Germany that we specified for exactly this workload — and we benchmarked language models strong enough for legal text but right-sized for the firm’s hardware. Which means: no oversized GPU bill for capability nobody uses.
Where it stands
Deliberately, the project is in its first phases: the pipeline is being built and tested stage by stage. Next, the same intelligence extends into the firm’s Outlook inbox — deadlines arriving by e-mail become tracked tasks in the tools the team already works in.
The real outcome
The strongest result isn’t in the system yet — it’s in the relationship. One automation question has become a long-term mandate: SANAD now acts as the firm’s standing technical partner for AI, automation and digital transformation. The first thing we built was trust; the pipeline is the proof of it.
Want to build something like this?
Tell us about your project. The first call is on us.