Summary:
This study designs and validates a protocol of LLM-enhanced benchmarking for the aerospace
industry, with an emphasis on launch-service providers. As commercial launches grow fiercely
competitive, firms need faster, deeper visibility into competitors’ technology choices, operating
practices, and cost structures. The research thus examines whether an advanced, generalpurpose,
decoder-only language model such as OpenAI’s o3, can efficiently and reliably support
the entire intelligence-gathering workflow.
Results:
Using two concrete examples, SpaceX’s in-sea recovery operations and the Chinese launch
services landscape, this study shows that a general purpose, decoder only LLM like OpenAI o3,
paired with web access and a clear research protocol leveraging iterative prompting that blends
LLM and non LLM sources, can make competitive benchmarking faster and more insightful, and
manageable by non-experts. Although designed in the context of aerospace engineering, the
protocol carries over to any industry with web coverage and can deliver real time savings and
deeper techno economic insight when backed by solid non LLM sources.
Summary:
This study focuses on enhancing customer satisfaction in contact centers by leveraging machine
learning (ML) algorithms to predict self-reported customer satisfaction scores, specifically the
Net Promoter Score (NPS), which measures customer loyalty and the likelihood of
recommending the service. Contact centers often struggle with gaining good visibility into
customer satisfaction with the service provided. This is because they typically rely on
satisfaction surveys that have low response rates, making it difficult to obtain a comprehensive
view. As a result, the visibility they do have may be biased by the subset of customers who
respond, and their actions remain largely reactive, responding only when and if customers
explicitly express dissatisfaction. By predicting detractors before they rate their experience
poorly, contact centers can implement targeted recovery strategies. Moreover, predictive
models like this can provide valuable insights for both short- and long-term improvements. In
the short term, such models can support recovery tactics, while in the long term, they can help
identify patterns in customer dissatisfaction to improve processes and guide agent training.
This study tests the application of three ML algorithms: KNN, Logistic Regression, and Random
Forests, to build a predictive model for NPS based on features available shortly after a phone
interaction. It also includes the implementation of one possible recovery tactic in a real-world
contact center scenario: remediation callback. The study’s main limitations are the restricted
availability of data for model development and the limited time available for testing the
selected recovery tactic.
Results:
The study’s results showed that the Random Forest algorithm outperformed other machine
learning models in predicting potential detractors. However, there is still significant potential
for improving model performance. Further model improvements, fine-tuning, and training on a
larger dataset could address limitations arising from the small representation of sentimentrelated
features available, which were identified as highly relevant in understanding customer
satisfaction. While the recovery tactic of empathetic, problem-solving callbacks showed a
positive impact on NPS, challenges remain in its implementation due to difficulties in reaching
customers after the initial interaction, which limits its overall effectiveness.
Summary:
This thesis investigates the integration of artificial intelligence (AI) into breast and lung cancer
screening programs in France, using Incepto Medical as a case study. It examines the current
state of national breast cancer screening, a mature program facing operational inefficiencies,
regional disparities, and delays, alongside lung cancer screening, which remains in the pilot
stage under the Impulsion pilot. Drawing from literature, workflow analysis through a Customer
Journey Map, stakeholder interviews, and a tailored Business Model Canvas, the work identifies
pain points, evaluates AI functionalities, and proposes integration scenarios: augmented
reading, assisted reporting, and digital coordination support. The analysis highlights the need
for AI adoption strategies tailored to each program’s maturity level, regulatory environment,
and clinical workflow, with an emphasis on interoperability, stakeholder engagement, and
clinician training.
Results:
The results show that the breast cancer screening program, although it has delivered significant
benefits to women’s health, now requires an intensive renewal to address persistent issues
such as variability in image interpretation, delays in result delivery, and inconsistencies in
reporting—challenges for which AI can provide targeted, incremental solutions, respecting the
program’s foundational strengths. For lung cancer screening, AI is already part of the Impulsion
pilot’s structure, but Incepto can enhance the current workflow through its comprehensive
approach to the patient pathway in screening, integrating detection, structured reporting, and
centralized data-monitoring into a cohesive system. Partnerships with institutional,
professional, and patient stakeholders, along with embedded radiologist training, are critical to
building trust and ensuring adoption. The thesis concludes that with a platform-based,
workflow-aligned approach, Incepto can position itself as an indispensable infrastructure
partner in national screening programs.
Summary:
This master thesis project focused on developing a discrete-event simulation model to support
decision-making in a complex industrial context. Using this approach, the study aimed to optimize
industrial operations and explore innovative solutions to enhance production efficiency.
The study employs a simulation software to model complex manufacturing workflows, evaluate
various scenarios, and optimize resource allocation. The methodology combines data-driven
analysis from enterprise systems with systematic scenario testing to identify optimal production
configurations.
The research demonstrates how advanced simulation techniques can support strategic decisionmaking
in high-precision manufacturing environments where traditional optimization
approaches may compromise product quality or operational flexibility.
Results:
The implementation of this flow simulation framework enables significant operational
improvements for the manufacturing workshop. Through systematic scenario testing and
resource optimization, the simulation demonstrates potential for substantial lead time
reductions while maintaining production quality standards.
The resulting simulation framework serves as a permanent strategic asset, enabling ongoing
operational decision-making through scenario testing capabilities. This tool allows manufacturing
teams to quickly evaluate the impact of production changes, resource adjustments, or layout
modifications before implementation, significantly reducing the risk associated with operational
changes in precision manufacturing environments.
Furthermore, the simulation methodology developed can be extended to analyze additional
workflows and production scenarios, providing a scalable foundation for continuous operational
improvement and strategic planning initiatives.
Summary:
The environmental performance of supply chains has become a central concern for companies
addressing sustainability challenges. In this context, the environmental impact of industrial
packaging is being thoroughly analysed, and initiatives are being launched to reimagine
processes in line with sustainability commitments. This thesis specifically analyses the
packaging used by suppliers to deliver mechanical tools to a watchmaking company. While this
packaging ensures the protection of tools as well as efficient distribution and in-house handling,
its environmental footprint raises concerns. Indeed, made entirely of plastic and not recycled
after use, the current packaging system is unsustainable over the long term.
This thesis aims to improve the existing packaging system for mechanical tools by developing a
more sustainable alternative. The first step involved mapping the current packaging flow to
gather data and insights, establishing the project’s scope. Building on this analysis and
addressing the specific constraints associated with this packaging, the development of a
sustainable process was subsequently initiated.
Results:
This thesis required numerous visits to the production site to closely observe the packaging
flow and gain an in-depth understanding of the logistical processes. Collaboration with various
stakeholders was essential. This included working with suppliers to quantify the volume of
packaging involved, partnering with the logistics department to use their expertise in designing
the improved process, and engaging with the Environmental Performance Department to
support the launch of the pilot phase.
The proposed solution was to transition from a single-use packaging system to a returnable
packaging system. The environmental benefits of this improvement were evaluated using the
Life Cycle Assessment (LCA) methodology, which demonstrated a theoretical reduction in
environmental impact by 85%. Practically, this solution involved establishing a process where
the packaging is collected directly from the production line, transported to a centralized stock,
inventoried, and redistributed to the suppliers.
Summary:
Advancements in surgical robotics have significantly improved precision, efficiency, and patient
outcomes. However, a well-organized servicing strategy is necessary to ensure the safe and
efficient operation of such complex medical devices. This project aims goal is to develop a
service model for Dynamis, a multi-arm robotic platform designed for spine surgery. Given the
regulatory challenges faced when entering the U.S market, a robust and scalable servicing
strategy is essential to guarantee device reliability, compliance and customer satisfaction.
To address this need, risk management was used to identify servicing challenge. Through the
application of Design Failure Mode and Effects Analysis (dFMEA) we identify potential failure
modes across the product subsystems. To guarantee an efficient and competitive service
model, industry best practices and competition benchmarks were also examined.
Results:
The results highlight how important preventive maintenance techniques are to avoid costly
downtime and improve operational reliability. Critical failure points were identified using
dFMEA, driving the creation of informed maintenance plan and service procedures. To support
these strategies, the Service Manual of the product was developed, providing standardized
maintenance guidelines, troubleshooting procedures, and installation protocols.
In addition to providing operational efficiency, the service model constitutes a strong
foundation for growth when Dynamis joins the U.S market and real-world data becomes
accessible. LEM Surgical can increase its servicing capacities in accordance with client and
regulatory requirements thanks to the organized framework.
By integrating risk management principles into servicing, this project contributes to enhancing
device reliability and participating to regulatory approval.