AI for law firms and companies with confidential documents
An AI assistant that searches case files, contracts and procedures and answers with a reference to the source can run on a server in the firm, without sending documents to outside services. Plus the infrastructure professional secrecy calls for: backups, access control and email that resists spoofing.
You are here: Services.
What usually gets in the way
The things I hear most often. If you recognise even two, it's worth talking.
- Public chatbots are tempting, but pasting client documents into them puts professional secrecy at risk.
- Finding a passage in hundreds of contracts and letters takes hours.
- Templates exist in many versions and nobody knows which one is current.
- People who shouldn't have access to documents do, or those who should don't.
- A ransomware attack or a disk failure can stop the work and expose client data.
What I can do
Each of these solutions can work on its own. We start with the one that gives the most.
Document search with sources
The assistant answers questions based on your documents and shows, with every answer, which document and passage it comes from. Permissions work like folders: everyone sees only their own matters.
AI implementationA language model on your server
Choosing hardware and a model for the size of the firm, installation and maintenance. The documents never leave the office.
AI implementationDrafts from your templates
The AI prepares a first version based on the current template and the case data, and a lawyer checks and signs it. The decision always rests with a person.
AI implementationSecure infrastructure
Ransomware-resistant backups with restore tests, two-factor sign-in, access control and email with SPF, DKIM and DMARC.
IT infrastructureWhat it can look like
A trainee asks: “Which lease agreements from the last two years include a penalty for late handover of the premises?” The assistant returns a list of agreements with quoted passages and links to the files, but only from matters the trainee has access to. Everything runs on a server in the firm. This is an example, not a deployment at a specific client.
Where to start
A suggested process. We'll fit the details to the company. How I work
- 01A conversation about the data
We agree which documents can go to the assistant, who has access to what and what the firm requires.
- 02A pilot on a slice
One type of document and a set of real questions with expected answers, so quality can be measured.
- 03Hardware and deployment
A server sized for the firm, installation, permissions and backups.
- 04Extending
More types of documents once the pilot answers well.
Questions and answers
Don't see your question? Write to me.
Do documents leave the firm?
No, if the model runs on your server. Then neither the documents nor the questions go to outside services.
Can the AI make mistakes?
Yes. That's why the assistant shows the source of every answer and a person checks what matters. We measure quality on a set of real questions before the assistant reaches the team.
What hardware is needed?
It depends on the number of people and the size of the model. Mostly what counts is the graphics card's memory. Details and a calculator are in the guide to hardware for a local model.
Is this compliant with the GDPR and professional secrecy?
Processing on your own server greatly reduces the risk, but the final assessment is the firm's, ideally with a data protection officer. I help describe where the data is and who has access to it.
Want to use AI without sending documents outside?
Write what kind of documents you work with. I'll tell you what can run on your own server.Services that help here
AI implementation on your own server
I run a language model on the company's own infrastructure: for searching documents, drafting replies and sorting requests.
More about the serviceIT infrastructure and server administration
Servers, backups, access control and monitoring.
More about the serviceCustom software and internal tools
Small apps and internal dashboards, for example for bookings, handling requests or keeping records.
More about the service