Internship - AI & Data Science (CASTLab)

Internship: Benchmarking and Optimizing AI Agent Configurations for Software Analysis

Duration: 6-month

Keywords: LLM/ Multi-agent systems/ Agent management

About CAST

CAST is the market leader in Software Intelligence. Its unique technology thoroughly examines the structure of complex software systems, with MRI-like precision. It delivers accurate, actionable, and automated views of software architecture, critical flaws, quality grades, sizing metrics, open-source usage, and cloud readiness levels. Hundreds of companies rely on CAST for greater objectivity in crucial business decisions, faster application modernization for Cloud, raising the quality and security of their custom software.

Founded more than 25 years ago and backed by nearly $200 million in R&D, CAST’s rapid analysis technology and its advanced ‘MRI for Software’ drive IT transformation and enable automation at the world’s largest systems integrators, hundreds of Global 2000 enterprises, and government agencies, in North America, Europe, India and in China.

CAST provides two sets of products of different technologies, implementation models, and usage:

• CAST Highlight is a SaaS product capable of performing rapid application portfolio analysis. It analyzes source code of applications to measure their cloud readiness, to analyze their composition, resiliency, and technical debt.
• CAST Imaging is an on-premises product (also cloud version since 2025) that can run for reverse-engineers all database structures, code components, and interdependencies in custom-built applications. It provides interactive and accurate architecture blueprints, data-call graphs, and end-to-end transaction flows.

Context

In AI workflows, tasks can be performed either by a single Large Language Model (LLM) or by a set of specialized agents working together (multi-agent systems). For example, one agent could extract dependencies, another validate licenses, and another generate summaries. However, the performance of such agents depends on their configuration (scope, tools, orchestration) and which tasks are suitable for multi-agent execution. Understanding these factors is key to building reliable and efficient AI-driven software analysis.

Proposed work

This project offers an exciting opportunity to explore how AI agents can be configured and optimized for software analysis tasks. Key milestones include:

• Experimenting with single vs. multi-agent setups and testing different configurations.
• Benchmarking tasks to see which benefit most from agent collaboration.
• Measuring performance across different LLMs (e.g., GPT vs. open-source models).
• Developing evaluation metrics and identifying best practices for agent orchestration.
• Building a lightweight knowledge base of viable configurations and insights for software analysis.

Students will have freedom to design experiments, explore creative setups, and uncover how agent behavior depends on task type and configuration.

Team:

Interns will be joining the CAST R&D team, a dynamic and innovative group of professionals specializing in software research and development. The team consists of experienced software engineers, data scientists, and industry experts who are passionate about pushing the boundaries of software technology. Working in this team offers:

- Exposure to cutting-edge research and development in software technology.

- Opportunities for mentorship and learning from seasoned professionals.

- A collaborative environment where creativity and innovation are encouraged.

- Involvement in projects that have a tangible impact on the industry.

Required Skills

To be successful in this internship, candidates should possess:

• A Bachelor’s degree (or currently pursuing) in Computer Science, Data Science, Engineering, or a related field.
• Strong analytical and problem-solving skills.
• Knowledge of Python and AI/ML libraries.
• Familiarity with LLMs and multi-agent systems is a plus.
• Interest in AI workflows, software analysis, and benchmarking experimental systems.
• Excellent problem-solving abilities and a keen attention to detail.
• Strong communication skills, both written and verbal, to effectively collaborate with the team and present findings.

Main Technology

The internship will involve working with various technologies, including but not limited to:

• Python & data science tools: Jupyter, Pandas, NumPy.
• Machine learning & AI libraries: scikit-learn, PyTorch, TensorFlow.
• LLM frameworks: OpenAI API, Hugging Face Transformers.
• Orchestration & evaluation tools: lightweight agents frameworks, experiment tracking (e.g., MLflow).
• Version control: Git/GitHub.

When: Flexible

Where:

Position is located at CAST’s France office at Meudon, ïle-de-France: 3 Rue Marcel Allégot, 92190 – Meudon

What we offer you

Lunch - Each employee benefits from a Swile card and access to FoodChéri

Remote - the possibility of remote work up to 3 days a week.

An exceptional working environment - we are well settled in a former mansion ideally located in Meudon with a beautiful garden (10 min by train from Montparnasse).

Feedback-Friendly Culture - In CAST we believe in effective feedback, we have been since day one normalizing feedback by incorporating it in our routine and creating safe space for employees to debate about what is and isn’t working.

Career prospects - In addition to our internal mobility policy that encourages employees to move between teams and subsidiaries.

At CAST, employees are encouraged to take on more and more responsibilities during their journey.

We are always looking for talented people who want to grow together with us. Would you like to join a truly entrepreneurial company and to be a part of our exciting journey? Apply today!