AI & automation
Useful AI, built into your environment
AI agents designed around your processes, with access rules, traceability and hosting matched to your data — rolled out progressively, from scoping to full-scale deployment.
AI · what it brings
AI agents that serve your processes
We design AI agents that understand your processes, work with your internal data and automate repetitive tasks — without replacing your existing tools.
Less manual entry
Automatic processing of incoming documents and requests.
Faster answers
Intelligent search (RAG) across your documents, emails and procedures.
Instant reports
Summaries and reports generated automatically.
Automatic tracking
Deadlines, follow-ups and approvals tracked without manual work.
Informed decisions
Support for drafting and decision-making.
Integrated with your tools
Microsoft 365, Teams, SharePoint, CRM, ERP.
AI · our method
From idea to secure AI agent
We scope, secure and orchestrate — technically, operationally and from a compliance standpoint.
Scope
High-value use cases, available data.
Design
Workflows, M365 / SharePoint / Azure integration.
Secure
Access, traceability, data separation.
Deploy
Proof of concept (POC) → pilot → progressive scale-up.
You keep control of your data
Access, sources, logs and data flows are defined at the design stage. Depending on the sensitivity of the project, we can use hosting in Europe, your Microsoft environment, or local inference options.
Client cases · AI
Real projects, not demos
Document processing automation
Client — automotive sector
Context
A daily flow of incoming documents from different suppliers, each with its own layout — processed manually, with slow data entry and a source of errors and delays.
Solution
An AI agent monitors a Microsoft 365 mailbox and analyses each document with Azure Document Intelligence: it recognises the document type and extracts the relevant fields even when the layout changes from one sender to another — with no hand-coded rule per template — then checks data consistency before publishing it to SharePoint.
Result
The flow is sorted, extracted and published to SharePoint automatically. Exceptions or inconsistencies are isolated for review, which avoids re-keying documents that are already correct.
Technologies used
Behind the scenes
Microsoft 365 mailbox
Document reading (OCR)
AI recognition — Azure Document Intelligence
Data extraction & validation
Publishing to SharePoint
Enterprise AI assistant on Microsoft 365 documentation
Client — enterprise Microsoft 365 environment
Context
Letting employees put AI to work on the company's internal documentation — not a generic chatbot, but an assistant able to find, analyse and summarise information with sourced answers, without exposing data to uncontrolled public services.
Solution
A RAG (Retrieval Augmented Generation) platform fully integrated into the client's Microsoft Azure and 365 environment. Every document added to SharePoint is automatically extracted, chunked and indexed; the assistant retrieves the passages that are genuinely relevant before passing them to the AI model, producing answers based on the company's own documents rather than the model's general knowledge alone. Access to the assistant is controlled via Microsoft Entra ID and limited to authorised users; access rules for indexed content are defined and tested within the application according to the project's scope.
Result
A private AI assistant that finds, analyses and summarises information in seconds — quotes, contracts, procedures or technical documents — with sources cited on every answer. Processing takes place within the Azure and Microsoft 365 environment defined for the solution.
Technologies used
Behind the scenes
Document added to SharePoint
Extraction & smart chunking
Vectorisation & indexing
Retrieving relevant passages (RAG)
Sourced answer to the user
A self-hosted AI agent to automate internal tasks
Client — services SME (self-hosted environment)
Context
Recurring internal tasks (follow-ups, reports, searching documentation) were taking up team time, and a consumer cloud AI tool wasn't an option: the client didn't want its internal exchanges passing through an uncontrolled third-party service.
Solution
Mensialis deployed a self-hosted instance of Hermes Agent, an open-source AI agent, on the client's infrastructure. Access, authorised tools and scheduled automations are defined within that scope. Reusable procedures are version-controlled as skills, reviewed and approved before going into production.
Result
Users submit their requests in natural language from the internal messaging tool. Business processing and application memory stay on the client's infrastructure; routing to the model depends on the inference architecture actually chosen for the project.
Technologies used
Behind the scenes
Request sent from internal messaging
Selecting the right skill (SKILL.md)
Execution by the agent, on-premise at the client
Context stored for future sessions
Answer or report returned to the user
Method
How we measure value
Every project is assessed against concrete criteria, not a general promise.
Volume processed
Manual time eliminated
Error rate
Processing time
Remaining human intervention rate
Let's spend 30 minutes scoping a use case
We'll look together at a process to automate — concrete, no obligation. With AI, we always start small: a measurable POC before scaling up.
Scope a use case