Skip to content
Division · AI assistants

AI that earns its place by saving time

Nuvaris offers AI assistants: AI assistants for document search, quotes, document extraction and repetitive tasks.

I build assistants confined to one real use case: search, extract, summarise or draft a reply.

AI assistants
The problem

You lose time searching through documents, entering invoices, preparing quotes or answering the same questions. AI can help, but only if the scope is clear.

What sets us apart

I may also conclude that AI is not the answer. If an FAQ, a form or simple automation is enough, we stop there.

In practice

An agent wired to your data — not a generic chatbot.

Knowledge base, business tools and human oversight: the agent answers with your company context, escalates when needed, and leaves an auditable trail.

Knowledge base & RAG
Assisted support & escalation
Common cases

What we can take over or build.

Each block starts from a concrete situation. Scoping then picks the right perimeter, without stacking up options nobody needs.

01

Document assistant (RAG)

Questions in plain language, answers sourced from your procedures, contracts or history.

Who it is for — Practices, engineering firms and document-heavy SMEs.

02

Quoting assistant

A first draft from your price list, with mandatory review before it goes out.

Who it is for — Tradespeople and providers with repetitive quotes.

03

Document extraction

Invoices and documents read automatically, data exported to your tools.

Who it is for — High-volume SMEs, accounting firms, logistics.

04

Support chatbot (site)

Answers drawn from your own information, handing over to a human when there is no answer.

Who it is for — Shops and services with frequently asked questions.

05

AI task automation

Email sorting, product sheets, summaries or drafts, always under supervision.

Who it is for — Small and medium businesses with repetitive tasks.

What we handle

What the work includes.

  • A measurable use case, chosen
  • An assistant wired into your documents or tools
  • Sourced answers wherever possible
  • Human approval for sensitive actions
  • A log of exchanges and tracked costs
  • A documented GDPR framework
Method

How we go about it.

  1. Choose

    We keep a single useful case: search, extraction, quoting or support.

  2. Test

    A pilot runs on a small volume of real data.

  3. Set the limits

    Boundaries, human approval, logging and GDPR rules.

  4. Track

    Answer quality, costs, errors and improvements over time.

Technologies

Multiple tools, multiple approaches.

Languages, frameworks, databases — and for AI, multiple models and LLMs. We choose for the need, without forcing a single stack.

  • Python
  • Node.js
  • TypeScript
  • PHP
  • Laravel
  • PostgreSQL
  • Redis
  • Docker
  • OpenAI
  • Claude (Anthropic)
  • Mistral
  • Gemini
  • Llama
  • LangChain
  • RAG · pgvector
  • OCR
Deliverables

What you get.

  • A scoped use case
  • An assistant connected to the data that matters
  • A simple interface
  • Guardrails and a log of exchanges
  • GDPR notes and retention rules
  • Cost and quality tracking
  • Short documentation
Whatever the practice

What never changes.

01

A written scope

Before any development, what the engagement covers fits on one page a non-technical reader can follow.

02

Decisions explained

Every structural decision is justified in writing — you can push back, and sometimes you will be right.

03

Acceptance testing before go-live

Nothing ships to production without a test environment where you sign it off yourself.

04

A handover of skills

By the end, someone on your side knows how to run the tool. That is a goal, not a bonus.

FAQ

Questions before starting.

If your question is not there, write to us: the answer will be just as direct.

Oui, comme tout outil. C'est pourquoi un humain valide toujours les actions sensibles (devis, e-mails) et l'assistant reste borné à un périmètre précis. Jamais d'action autonome risquée.

On encadre tout par un contrat de sous-traitance (DPA), on minimise les données envoyées, et pour les données sensibles on peut utiliser un modèle hébergé en UE (Mistral). Hébergement européen documenté.

Le coût de traitement est faible (souvent quelques euros à quelques dizaines par mois pour une TPE). On pose toujours un plafond d'usage pour éviter les dérives, en toute transparence.

Pas toujours — et on vous le dira franchement. Pour un site à faible trafic, une bonne FAQ suffit ; pour la prise de rendez-vous, un simple agenda en ligne fait le job. On ne déploie l'IA que là où elle fait gagner du temps mesurable.

Go into detail

Something around AI assistants?

Describe your situation. We reply with a first technical read and a possible path forward.