Malo Lecoursonnais

Context: Participate in the security strategy for the use of LLMs by the Orange group. Elaboration of a state of the art on the subject, exploration of existing solutions, benchmarks and analysis of results. Competitive technological watch.

Skills:

AI, LLMs (large language models), cybersecurity, risk analysis, vulnerability analysis, red teaming attacks, Github, Python, deployment, open-source, business strategy analysis

State of the art: securing LLM usage by Orange

- Contextualize and frame the subject

List vulnerabilities

Risk assessment

History and significant past attacks

- Exploration of existing solutions

Approach subject matter experts (LinkedIn, professional email, internal experts)

Inventory of solutions with their business plan

Progress report, ongoing communication with my team's expert

- Deployment of solutions & benchmarks

Integration of open-source solutions via Github

Deployment in the Orange environment

Interconnection of LLM APIs made in Orange

Tests at scale

- Analysis of results

Analysis of results, consistency

Draw conclusions from the data

Compare each solution based on models

- Participation in Orange's AI security strategy

Writing key strategy report

Discussion of short/medium term plan for Orange AI security projects

Creation of educational modules and training

- Creation of training materials

- Adapt courses to the students' level and available resources

- Live in a different culture

Within Orange Innovation, in a chatbot and AI solutions team for France and the MEA region.

- Product and project management: Technical and functional referent for 3 major clients (MEA). Management of the service's scaling, from 3M to 9M annual requests.

Product Management: Steering the product roadmap and backlog prioritization (Jira, Trello).

Delivery & Tech: Realization of 4 GenAI Agent PoCs and development of associated APIs (Python, Node.js) in a containerized environment (Docker, Kubernetes).

Support: Training of 5 international teams on new Generative AI and Low Code solutions.

AI Security: Contribution to LLM security (ChatGPT, Mistral), risk analysis (prompt injection), and evaluation of red-teaming solutions.