Systems
A system here is something the lab built and runs with a client, inside that client's environment.
Current system
Novalis
Novalis answers questions about an organization's own documents, with every claim tied to a source passage.
Most organizations keep their knowledge in piles of documents that nobody can search: policies, reports, scans, email. Novalis indexes the pile and answers questions about it in plain language. Every claim links back to the passage it came from, so readers can check instead of trust. When the documents do not contain the answer, Novalis says so.
It answers only from what it retrieves, scans incoming documents for injected instructions, and measures its own answer quality on your real questions. It runs inside your infrastructure and, by default, calls no outside services.
- Your existing documents: office files, PDFs, scans, spreadsheets, email. More than 25 file types.
- Plain-language answers, every claim cited to its source. It says when it does not know.
- Inside your infrastructure. No outside calls by default.
- Separate knowledge bases per team or topic, with access scoped per person.
- Your team runs it from an operator panel. No shell required.
In progress
More systems are in progress. We add one when the research behind it is ready to run somewhere real.
Engagements
Four stages, and the same small team stays with the work through all of them.
Scoping
A few weeks looking at your documents, your questions, and your environment. If the fit is wrong, we say so and stop.
Build
We build against your data from day one. You see working software early, and a shared evaluation set shows how well it is doing.
Deployment
The system deploys inside your infrastructure, under your security review. Your team learns to run it while we are still around.
After handover
We stay involved: model upgrades, new document types, the occasional hard problem. The system keeps working when we leave.
To talk about an engagement, write to hello@mnemiclabs.com.