Internship - IT & Architecture Brain

Location: Paris, Ile-de-France, France Department: Enterprise Architecture

The intern will design and build a proof-of-concept “IT & Architecture Brain,” structured not as one monolithic assistant but a set of specialized agents, each responsible for one activity, each accumulating its own working context and knowledge over time, all drawing on a shared transverse knowledge layer so they stay consistent with one another.

Architecture knowledge agent — answers “what exists, why was it built this way, and how does it connect” by retrieving and reasoning over the documentations, ADRs.

Change impact agent — reasons about blast radius before a change (“if we deprecate this API/service, what breaks and who owns it”), useful for architecture and change advisory boards.

Incident & runbook agent — holds persistent memory across past incidents and postmortems, surfacing “this looks like the outage from [date], here’s what fixed it” instead of starting from zero each time.

Governance & compliance agent — checks new designs/ADRs against existing standards, reference architectures, and prior decisions, flagging inconsistencies or duplicated capabilities across teams.

Rather than four disconnected tools, the goal is to demonstrate that one well-structured mesh, sitting on one shared knowledge layer, can power all four agents.

1)Discovery : Review of enterprise brain tooling approaches; inventory of available sources; scope confirmation

2)Shared layer & mesh design : Define the shared data model (systems, dependencies, decisions, incidents); define each agent’s scope, context, and memory model; sketch the harness (routing, handoff, access control); design evaluation metrics

3)Core build  EA knowledge agent : Ingest system catalog, ADRs ; build retrieval + reasoning for agent 1 ;  build a minimal harness (single-agent routing, logging)

4)Mesh extension : Add one or two more agents (change impact, incident memory, or governance); extend the harness to multi-agent routing and handoff; wire agents to the shared layer and their own memory

5)Reliability & evaluation : Consistency checks across agents, access control, audit logging, fallback/escalation behavior, benchmark against stateless baseline, validation with architects/engineers

Documentation : Final report, demo, recommendations for scaling the mesh to more agents/teams

A working prototype of the agent mesh (codebase + demo): at least two agents, a shared knowledge layer, and a harness coordinating them

An architecture document describing the shared data model, each agent’s scope/context/memory, and the harness’s routing and access-control logic

An evaluation report covering per-agent accuracy

A final presentation with recommendations on extending the mesh to additional agents, data sources, or teams

Required experience & competencies

Programming skills (Python preferred)

Familiarity with LLMs and how they’re used

Basic understanding of information retrieval concepts (embeddings, vector search)

Interest in software/enterprise architecture concepts (system design, dependencies, documentation practices)

Comfort working with ambiguous, evolving requirements and iterating quickly

Not required but nice to have

Experience with RAG frameworks (LangChain, or similar)

Exposure to graph databases or knowledge graphs (Neo4j, etc.)

Familiarity with multi-agent frameworks, orchestration patterns

Interest in distributed systems reliability concepts (circuit breakers, fallback, observability) applied to AI agents