Services · Generative AI services
Generative AI that answers from your documents — inside your walls
Assistants, document automation and LLM features built on your own knowledge, with citations, access control and real evaluation — running on-premise or EU-hosted so confidentiality is an architecture decision, not a risk you accept.
When to call us
Sound familiar?
A demo on ten documents is not a system on ten thousand, with permissions, updates and a quality bar.
Shadow AI is already your generative-AI strategy — just not one you chose.
Contracts, procedures, reports, emails — searchable by nobody, answerable by no one in under a day.
What you get
Outcomes, not deliverables first
Retrieval-augmented generation over your own documents: every answer points to its source, and “I don’t know” is a valid output.
Open-weight models on your hardware or EU-hosted inference, per-user access control, no training on your data.
Evaluation sets, hallucination and citation-accuracy rates, guardrails — numbers your risk committee can read.
What we deliver
The scope you can actually sign
01
Use-case scoping & feasibility
Which questions, which documents, which users — and whether the data can support the answer quality you need.
02
Document ingestion pipelines
PDF, DOCX, email, SharePoint, DMS — parsed, chunked, enriched and kept in sync, with permissions preserved.
03
RAG assistants with citations & access control
Chat or embedded Q&A over your knowledge, answers with sources, each user seeing only what they are allowed to.
04
LLM features in your workflows
Drafting in house style, classification, extraction, summarisation — integrated where work happens, not in yet another tab.
05
Evaluation harness & guardrails
Test sets from your real questions, automatic scoring, prompt-injection and data-leak defences, monitoring in production.
06
Operation & cost control
Model routing, caching, usage dashboards, and the runbook your team needs to own it.
How it runs
Fixed scope, fixed price per phase — you decide at each step.
Scope (1 week)
Questions, documents, users, constraints, quality bar — and a fixed price for the pilot.
Pilot on your documents (3–4 weeks)
A working assistant on a real corpus, evaluated against real questions, on your infrastructure.
Production (phased)
Integrations, access control, monitoring, roll-out by team.
Operate & extend
New sources, new use cases, model upgrades — with the evaluation harness guarding quality.
Regulation & sovereignty
Generative AI under the AI Act and GDPR
Most internal assistants are limited-risk under the AI Act — transparency to users and AI-literacy duties apply, and AI-generated content destined for the public must be labelled. The real exposure is GDPR: prompts, retrieved passages and logs are personal data the moment a client or employee appears in them, which is why we default to on-premise or EU-hosted inference, access control at retrieval time and minimal logging.
Questions
Which models do you use?
We are model-agnostic: open-weight models (Llama, Mistral, Qwen families) on your hardware or EU hosting when confidentiality rules, API models through EU regions when the data allows it — chosen by evaluation on your documents, not by fashion.
What about hallucinations?
Grounding in your documents with citations, answer-refusal when retrieval is weak, and an evaluation set that measures citation accuracy before and after every change. We report the numbers; we do not promise zero.
Can it connect to SharePoint, Teams or our DMS?
Yes — connectors for SharePoint/OneDrive, network shares, email, common DMS and ticketing tools, with permission-aware sync. Exotic systems get a custom connector.
Is this the same as KnowledgeRAG?
KnowledgeRAG is our product — the fastest path to a cited assistant over your documents. These services cover everything around and beyond it: scoping, pipelines, custom features, evaluation and operation, with KnowledgeRAG or with your own stack.
Start with a 30-minute conversation
Tell us the problem, not a spec. You get an honest read on feasibility, data, compliance exposure and a first step — within one business day.