Service 02
Your company knows more than it can find.
Most company knowledge sits in unstructured files nobody can retrieve. We classify it, cut the ROT, and build the retrieval, RAG, and workflow layer that turns stored data into working knowledge.
Who this is for
Three situations we see every week
Teams drowning in their own files
Decades of documents, three generations of folder structures, and nobody can find last year’s contract. Search fails because the data underneath is a mess.
Companies whose AI pilots stalled
The chatbot demo was great, then it hallucinated in front of a client. AI is only as reliable as the data and guardrails underneath it.
Firms invisible to AI search
Clients ask ChatGPT and Perplexity for recommendations. If those engines cannot read you, you are not in the answer.
What we build
Data classification & ROT cleanup
We map what you have, label it, and remove the redundant, obsolete, and trivial. The result: less risk, lower cost, and data AI can actually work with.
Retrieval & RAG pipelines
Search that understands meaning, not just keywords. AI answers grounded in your own documents, with sources your team can verify.
AI workflows
Repetitive knowledge work, automated: document processing, report drafting, triage, and the glue between your tools.
Internal AI assistants
A company copilot that answers from your own knowledge base. Your team asks in plain language, gets grounded answers with sources, and stops digging through folders.
Document intelligence & intake automation
Contracts, invoices, applications, and inbox triage turned into structured data. Extraction pipelines that cut processing time from hours to minutes.
GEO & machine-visibility audits
How do ChatGPT, Perplexity, and Google AI describe you today? We measure it, then fix it: schema, llms.txt, and content structure that machines quote.
AI governance & compliance readiness
Access rules, audit trails, and deployment policy aligned with the Swiss nLPD and the EU AI Act. AI your lawyers and your clients can live with.
Evaluation & monitoring
Ongoing evals, hallucination and drift checks, and cost monitoring for the AI systems you run. Reliability is not a launch feature; it is a practice.
Machine legibility
Structured data, clean schemas, and content formats that AI search engines can read, quote, and recommend. Your expertise, findable by the machines your clients ask.
How we work
Audit. Build. Run.
01
Audit
We map your data estate: what exists, where it lives, who can access it, and what share of it is redundant, obsolete, or trivial. You get an honest picture before anything gets built.
Data map and prioritized roadmap
02
Build
Classification, cleanup, then the systems on top: retrieval and RAG pipelines, internal assistants, document intelligence, and the workflows that connect your tools.
Working systems on your own data
03
Run
Evals, hallucination and drift monitoring, cost tracking, and governance. We keep the systems reliable as your data, team, and the models themselves change.
Monthly reliability and impact report
Ways to work together
Audit first. Build what the audit justifies. Keep it reliable.
Fixed scope
Data Audit Sprint
A fixed-scope assessment of your data estate and AI readiness, ending in a prioritized roadmap. Useful on its own, and the foundation for everything after.
Project based
Build Project
A defined system, delivered: a RAG pipeline, an internal assistant, an intake automation. Scoped, built, tested, and handed over working.
Monthly retainer
Run Retainer
Ongoing evaluation, monitoring, and governance of the systems in production. The recurring layer that keeps AI trustworthy after launch.
Questions companies ask us
What is ROT data and why does it matter? +
ROT stands for redundant, obsolete, and trivial data. It typically makes up the majority of company storage. It slows retrieval, poisons AI outputs, raises storage cost, and creates compliance risk. Cleaning it is the first step toward reliable AI systems.
What does a RAG pipeline do for my company? +
Retrieval-augmented generation lets AI answer questions using your own documents instead of guessing. Your team gets accurate answers grounded in your data, with sources, instead of generic model output.
We already tried an AI tool and it disappointed. Why would this be different? +
Most AI disappointments are data problems wearing an AI costume. Tools bolted onto messy, unclassified data produce messy, unreliable answers. We fix the foundation first, then build on it, and we monitor what we ship.
Do we need content strategy to use this service? +
No. The services stand alone. But they reinforce each other: clean infrastructure surfaces insights worth publishing, and published expertise attracts clients who need infrastructure.