
SERVICE 04 / 05 · ASK · AI ASSISTANTS
THE PROBLEM, IN YOUR WORDS
“Answersareburiedindocuments;reportstakedays.”
AI assistants
We build assistants that answer from your own documents and cite the page, or decline when nothing covers the question.
DELIVERABLES
What you get
All industries
- 01Document assistants that cite the page, or declineAnswers drawn from your documents with the page they come from, or a plain refusal when nothing covers the question.
- 02Report-drafting agentsAgents gather the evidence and draft the report. A person approves it before it goes out.
- 03Evaluation on real questionsA test set scored for correct answers, correct citations and correct refusals, before launch and after.
- 04English and FrenchQuestions and answers in both languages.
What is the economic threshold for soybean aphid?
SEARCHING YOUR DOCUMENTS
HOW WE APPROACH IT
- 01Collect documents and real questionsThe documents people rely on, and the questions they actually ask.
- 02Build and test against those questionsWe measure answers, citations and refusals before anyone relies on the assistant.
- 03Deploy with logging and feedbackEvery answer is logged with its sources. Users flag mistakes and the test set grows.

TECHNOLOGIES WE COMMONLY USE
We pick tools for your environment, not ours.
- Python
- Retrieval with citations
- Vector search
- pgvector
- Hosted or open-weight LLMs
- Evaluation sets
- FastAPI
- Docker
SOLUTIONS THAT USE THIS SERVICE
What it delivers
Questions
What happens when the documents don’t cover a question?
The assistant says so and does not guess. Those questions are logged, so gaps in your documents become visible.
Can it run on our own servers?
Yes. Where data cannot leave your network, we use models that run on your own infrastructure.
Does it work in French?
Yes. Questions and answers work in English and French.
Can this be funded?
Some Canadian programs, such as NRC IRAP, may cover parts of an AI project; we can help you scope it.
Sound familiar? Tell us where you are today.
Next service: Data engineering, MLOps and DevOps

