BTP / Ingénierie

Jarvis

Multi-Agent Mixture of Experts

Jarvis project — Multi-Agent Mixture of Experts
Multi-AgentsRAGOCRWorkflowsVoice

Key result

4 agents

domain-specific, in production: landscaping, civil works, hydraulics, specifications

Chapter 1

Faster BTP/VRD studies, without sacrificing rigour.

The client wanted to speed up and secure its landscaping and VRD studies (utilities, hydraulics, estimating, technical documents), on both public and private projects.

The challenge: quickly find the right regulatory information, exploit plans and diagrams, and produce consistent deliverables.

The context

An engineering consultancy, dense files, little time.

1 sector · construction & public engineering

Chapter 2

A team of AI agents, one per specialty.

We designed a multi-agent architecture orchestrated by workflows. Each agent is backed by a dedicated knowledge base and a hybrid OCR + text model to process documents, plans and diagrams.

The V1 agents

Landscaping agent
Plant species, landscaping standards, site integration.
VRD agent
Utilities, roadways, connections.
Hydraulics agent
Sizing, flow calculations, standards.
CCTP agent
Drafting of consistent technical documents.
Chapter 3

A multimodal stack, spoken and written.

Simple interface, secure access, text or voice input, text or audio replies, PDF / DOC / standard-format exports.

Everything is designed for on-site use — not just the office.

The stack

RAGHybrid OCR + textMulti-agent workflowsVoice in/outDOC/PDF exportsAuth
Chapter 4

A V1 delivered, a roadmap that follows.

The V1 is in production. The roadmap includes additional agents (estimating, compliance), and deeper integration into the firm's existing tools.

As always, we measure the real gains in hours per file — not in promises.

The result

Agents in production
4 specialties
Supported modalities
Text · Voice · Plans
Status
V1 delivered

Along the same lines

A similar project?

Let's talk about it.

30 minutes to qualify your need and see how we can support you — no commitment.

Discuss a similar project