Software Engineering

Improving the understanding of needs described in natural language and enabling the machine to design and develop IT solutions based on natural language descriptions is one of the missions we have set ourselves at Novelis

In our daily life, AI is present in many fields, from medicine to automotive and industry… AI itself is a very vast field gathering several subfields. In recent years we have seen great advances in AI and software engineering.  

Software engineering is the science of industrial engineering which allows to study methods and good practices that engineers must follow to create quality software by minimizing the time to market.

The alliance between AI and software engineering 

One of the biggest advances in recent years in the NLP (Natural Language Processing) subfield of AI are the Transformer models.   

Transformers (self-attentive) is a deep learning model introduced in 2017 for modeling natural language. It allows a machine (computer) to understand, classify and generate natural language text. Transformers uses an attention mechanism that allows to process data in parellel faster than other approaches used before in natural language processing (case of recurrent neural networks: RNN).   

The Transformer can be used in different cases: translation, text summary, text generation, identification of named entities (places, cities, people’s names, ….), as it can be used in Question/Answer type tasks.

At Novelis we use this model to illustrate the understanding of a specification, which earned us the first place in the international challenge organized by Microsoft: the CodeXGlue challenge.   

Improving the understanding of needs described in natural language, giving the machine the possibility to design and develop computing solutions, algorithms, parts of applications based on natural language description represents one of our biggest research works. 

We want to advance the science of artificial developers that can assist human developers in their daily development tasks. The goal is to remove low value-added tasks for developers to allow them to refocus on tasks requiring human appreciation, more complex. It will never be a question of replacing them, but of assisting them.   

In this sense, we are working on a program to design and implement artificial developers: AIDA

Another example of the application of AI in software construction within Novelis is image processing: we are able to generate the graphical interfaces of applications from images describing the screens of these applications. We are working on the implementation of an AI that will generate the computer code to create the HMI (Human Machine Interface).

Novelis Research Lab

At Novelis, we are committed to using new technologies as tools to better serve our clients’ business challenges and thus better support them in their transformation.

To meet these needs, we have set up an ambitious laboratory in terms of research and development with substantial investments: we invest more than a quarter of our turnover in research.
This R&D laboratory is housed at the École Polytechnique and benefits from the school’s scientific ecosystem. A dozen PhD researchers work there on a daily basis on fundamental and experimental research on AI: machine learning, image processing and NLP.

Our scientific publications

Low-Cost Language Models: Survey and Performance Evaluation on Python Code Generation

Discover the first version of our scientific publication "Low-cost deep language models: Survey and...

March digest – Summary of our Novelis Research posts on AI in Time Series Forecasting

At Novelis, we are committed to using new technologies as tools to better respond to our customers'...

October digest – Recap of our Novelis Research posts about on language modeling technologies (LLM)

At Novelis, we are committed to using new technologies as tools to better respond to our customers'...

September digest – Recap of our Novelis Research posts about Computer Vision

At Novelis, we are committed to using new technologies as tools to better respond to our customers'...

GPT-3.5 vs GPT-4: Evaluating ChatGPT’s Reasoning Performance in Zero-shot Learning

Discover the first version of our scientific publication "GPT-3.5 vs GPT-4: Evaluating ChatGPT's Re...

All scientific publications

They trust us