Services · AI and ML
AI that does real work
Most AI projects stall somewhere between the demo and the second month. We build the unglamorous parts that get them past it: access to the right data, permissions, evaluation, cost limits, and a team that is still there when something needs fixing.
Running in production today
-
Knowledge assistants
Staff ask questions in plain language and get answers from your own documents, with links to the sources. We built a SharePoint indexer for exactly this, and it runs in production today; the same approach works for shared drives, wikis and databases.
-
Private AI platforms
One gateway to more than 200 models, running in the client's own cloud, with per-user budgets and audit trails. It is what we run for public- and private-sector clients as VAIOS.
-
AI inside your product
Search that understands questions, summaries, drafting and copilots, built into the apps your customers already use rather than bolted on beside them.
-
Machine learning on messy data
Classification, matching and extraction where the data is inconsistent, multilingual or half-migrated. We have tagged candidates, jobs and skills across two legacy systems at once.
What we can build next
-
MCP servers and agents
Let Claude, ChatGPT or your own agents work with one system at a time, such as looking up an order in your ERP or drafting a reply in your CRM. They act with the same permissions as the person asking, every action is logged, and anything that changes data waits for approval. Built on the Model Context Protocol, the open standard for connecting AI assistants to software.
-
Evaluation and guardrails
Test sets, automatic evaluations and regression checks on every prompt and model change, plus cost and latency budgets, so 'it seems better' becomes a number you can track.
How it starts
One assistant, one data source
Start with one assistant on one data source. We test it against real questions from your team before anyone relies on it.
- Claude
- OpenAI
- Gemini
- Open WebUI
- LiteLLM
- Langflow
- Python
AI terms, plainly
The words on this page, without the hype.
- RAG (retrieval-augmented generation)
- The model looks things up in your documents before it answers, and shows you where the answer came from.
- MCP (Model Context Protocol)
- An open standard for connecting AI assistants to tools and data. Build one MCP server for a system and any compatible assistant can use it.
- Agent
- An AI that can take steps on its own: look something up, call a system, check the result, within limits you set.
- Evals
- Automated tests for AI output. They tell you whether a change made answers better or worse before your users find out.
- Private deployment
- Models run through your own cloud account under enterprise terms, so your data is not used for training and stays where you decide.
- Guardrails
- Rules around the model: what it may see, what it may do, and what gets blocked or redacted on the way in and out.
Start with one assistant.
Tell us which documents or system your team keeps asking questions about. We will say what a first assistant on it would take.
Start with one assistant