Designing an internal AI platform

8modular components
~€1,000per month excl. LLM
€90base infra per month
3usage levels

01 · SITUATION

The context 

Off-the-shelf AI solutions are vendor-branded, high per-request cost, and do not belong to the firm. The make-or-buy arbitration had never been formalized or costed.

02 · METHOD

8-component modular architecture 

Needs mapping

Identification of 3 usage levels: simple agents, complex agents, internal AI school.

8-component architecture

Interface, orchestrator backend, storage/DB, LLM & RAG, toolbox, copilot, cloud & rights, interconnected via MCP/API.

Make-or-buy arbitration

TCO costing per component, comparison with vendor solutions (Konverso and equivalents), documented recommendation.

Internal deployment

Deployment on existing infrastructure. Free choice of model (including sovereign). Independence from any single vendor.

ARCHITECTURE DIAGRAM

Target platform architecture, modular building blocks interconnected via MCP/API (working document).

Target platform architecture, modular building blocks interconnected via MCP/API (working document).

03 · RESULTS

Architecture delivered 

Platform costed at ~€1,000/month excl. LLM (€90 base infrastructure)
Independence from any single vendor
Free model choice (including sovereign models)
Own asset usable as a commercial argument
Extensible architecture for all 3 usage levels

Next case

Training & enabling teams on generative AI

An internal AI platform project?

Let's book 30 minutes.

dianejmaurin@gmail.comLinkedInMalt
Designing an internal AI platform — Diane Maurin