Before
Engineers had to move between separate tools to cross-check topology, inventory, services and capacity.
Available from September 2026 France, open to international opportunities
AI Engineer, agents and generative AI
I build AI tools that work in the real world.
An ESIEE Paris engineering graduate, I enjoy starting with a concrete problem and building the right tool for it. My work combines generative AI, business data and software development.
Network Q&A
topology, inventory, services and capacity
2,172
optical services available through controlled tools
1 workflow
from a feasibility request to a reviewable report
My focus
AI agents→Business tools
Case study at Orange Innovation
Andiamo Agent
I designed Andiamo, a conversational assistant for querying Orange optical-network data in natural language. A feasibility request automatically triggers the complete engineering analysis and report generation.
With Andiamo
One conversation answers everyday questions and launches the full feasibility workflow when needed.
AN
Online Andiamo
How the assistant is used
Ask in natural language
“Which services use this network link?”
“Can this new optical connection be deployed?”
Andiamo identifies the intent
01
Network answer
Topology · Inventory · Services · Capacity
02
Feasibility workflow
Route · Equipment · Signal · Report
My contribution
From architecture to a demonstrable product.
- 01
Design
Conversational router and multi-agent orchestration.
LangGraphArchitecture - 02
Connect
Access to network data and engineering tools.
MCPGraphQLPostgreSQL - 03
Make reliable
Validation, traceability and structured outputs.
ControlsQuality - 04
Deliver
Demonstration interface and automated reports.
UIReporting
2,172
network services
accessible through controlled agent tools
< 6 s
median latency
measured in the demonstration environment
4 analyses
combined in one study
route, capacity, equipment and signal quality
1 report
ready for review
including the decision, supporting data and simulation results
Demos
See the workflow in action.
Short extracts from an internal engineering prototype. All visible data has been anonymized.
01 / Network map
Explore the topology
Interactive view of the links, regional filters and remaining capacity on each connection.
What to notice: the map exposes possible routes and available capacity before the analysis starts.
02 / AI assistant
Query the network
A business question written in natural language is routed to the right agents and data sources.
What to notice: the question is interpreted, then assigned to the appropriate data and engineering tools.
03 / End-to-end workflow
Run a feasibility study
Route selection, equipment choice, signal-quality simulation and generation of a decision report.
What to notice: all four analyses are combined in a report that supports the final decision.
Architecture
One assistant, two response paths.
The router selects either a direct data answer or a verified engineering workflow.
From a network question to the right level of analysis
01
→ Network question
Topology, inventory, services, capacity or feasibility
02
→ Router and agents
Network data Engineering tools GNPy simulation
Network answer
Topology · Inventory · Services · Capacity
Feasibility workflow
Route · Equipment · Signal · Report
Reliability by design
The model coordinates. Engineering tools decide.
The LLM coordinates approved tools; it does not perform the engineering calculations itself.
Controlled data access
Agents can only access approved tools and data sources.
Deterministic calculations
Engineering calculations are delegated to specialized Python tools and GNPy.
Traceable outputs
The final report exposes the route, capacity, equipment and simulation results supporting the decision.
Key engineering decisions
The main technical choices and the problem that required the most care.
Why LangGraph? +
I first tested n8n, then selected LangGraph because the project required more modular agent state management, conditional routing and custom validation logic.
Why MCP? +
MCP provided a consistent interface between the agents and heterogeneous tools such as databases, APIs and simulation software.
What was hardest? +
The main challenge was not generating text, but providing agents with reliable, up-to-date operational data and ensuring that every result remained technically verifiable.
Generated report
The final decision, with the evidence needed to review it.
The English report summarizes the selected route, available capacity, required equipment and signal-quality simulation. The original detailed report remains available in French.
About
What interests me about AI.
I recently graduated from ESIEE Paris in Data Science and AI. I am particularly interested in AI agents: specialized assistants that work together and use tools.
At Orange Innovation, I learned to put these ideas to work in a real environment, with imperfect data, technical constraints and users who need clear answers.
Product
Start with the business need, not the model.
Reliability
Measure, trace and validate responses.
Systems
Connect AI models to real tools and data.
Career
Experience
My experience spans research, product development and client-facing work.
02/2026 to present
Final-year internship
01Orange Innovation
AI Engineer for optical network management
- Designed a multi-agent assistant to analyze and simulate a long-haul optical network from questions written in natural language.
- Connected the AI system to network maps, available equipment and active services through GraphQL, PostgreSQL and Python tools.
- Validated the full workflow, from the user’s question to a report detailing the route, capacity, equipment and signal quality.
05/2025 to 08/2025
Research internship
02- Compared sociological biases and behavior across five language models, including LLaMA and GPT, on human and societal topics.
- Designed a LangGraph multi-agent workflow to analyze around thirty interviews using the Braun and Clarke thematic analysis method and the Groq API.
- Added inter-agent validation and consistency checks to improve the reliability of thematic coding.
10/2025 to 12/2025
Academic project · 11 weeks
03Capgemini Agile Testing Chair × ESIEE Paris
Product Owner, SaaS development project
Discover the Capgemini Chair↗- Led the product side of a SaaS built by a project team around a client need, from requirement clarification through delivery.
- Defined and prioritized the backlog, wrote user stories and tracked progress with the development team in Jira.
- Took part in sprints and Scrum ceremonies while applying development, testing, continuous integration and DevOps practices.
01/2025 to 02/2026
04- Managed more than five client engagements on technology topics, from requirements analysis to final delivery.
- Managed budgets ranging from €1,000 to €20,000 depending on project complexity.
- Prepared business proposals and client presentations while coordinating teams of student consultants.
Skills
The tools I use.
I mainly work with AI agents, data and software development tools.
AI agents and generative AI
01Build assistants that collaborate, use tools and verify their answers.
LangGraphLangChainMCPRAGRequest routingTool-using agents
Software and data
02Connect AI systems to databases and reliable services.
PythonFastAPIGraphQLPostgreSQLRedisSQL
Machine learning
03Train, compare and evaluate models with a rigorous approach.
PyTorchTensorFlowScikit-learnXGBoostPandasEvaluation
Production
04Turn a prototype into a monitored, tested and deployable application.
DockerKubernetesCI/CDGitLinuxMonitoring
In simple terms: an LLM understands and generates text; RAG helps it retrieve reliable information; MCP provides a standard way to use external tools.
Selected work
Other projects
Kaggle project, 04/2025 to 05/2025
Building energy consumption prediction
Data analysis, feature engineering, rigorous comparison of several models and fine-tuning of the two best performers.
PythonXGBoostRandom ForestPandasScikit-learn
Academic project, 11/2024 to 12/2024
Data collection, storage and visualization pipeline
An end-to-end application that collects web data, stores it, speeds up queries, monitors quality and displays the results in a dashboard.
PythonScrapyMongoDBRedisKibanaDash
Contact
Working on an AI project?
Let’s talk.
I am looking for an AI Engineer or Generative AI Engineer position from September 2026. I am mobile across France and open to international opportunities.