From Legacy to future: ALTEN’s AI approach to code modernisation

What was the original challenge you saw that inspired this platform? 

Roel: We noticed that many of our clients face challenges when it comes to using AI and managing legacy code. Code modernisation has long been part of our expertise, and now we have developed tools to address these issues more effectively.  

We start by helping clients understand their code: with an LLM (Large Language Model), they can ask questions directly about the codebase, and with documentation generation, we deliver clear, structured documentation in their preferred format. On top of that, we use AI-assisted coding tools to accelerate development with AI-assisted code completion and smarter code navigation. 

How does the ALTEN platform use AI to understand complex code bases?     

Roel: Currently, our platform has three main tools:  LLMs for developers (a Copilot-like assistant), a Codebase Q&A system, and a Documentation Generator to help teams better understand their codebase.  

Arthur: The Artificial Intelligence is not only used to simplify complex databases, its main objective is to streamline the creation of documentation for legacy code. By legacy code, we refer to projects within companies where the passage of time or employee turnover has led to a loss of expertise and knowledge.  

Valentin: Our platform supports this by using AI to answer questions from the user. In practice, it lets the user extract precise information from long and complex documents without the need to read everything. 

Clément: The LLM uses tools to access the code base. It can retrieve keywords from multiple files and provide contextually accurate responses to the user. Additionally, the LLM is effective in constructing sentences and engaging in conversations with the user to collaboratively complete documentation. 

What makes this ALTEN tool different from off-the-shelf AI products?    

Valentin: Naturally, the goal of this platform is to function like a Virtual Assistant. 

Clément: Our key focus on ALTEN’s side is to store data locally to prevent potential online data leaks, with the LLM being the only external connection. 

Arthur: Our AI tool is directly connected to the project itself, giving it the ability to analyse the codebase and operate with greater autonomy. We designed the tool to follow a structured process, requesting user confirmation at each critical step to ensure accuracy and alignment. This approach makes our solution a truly custom-made experience for every client. 

Roel: What sets our platform apart is its direct focus on the client’s codebase, combined with software architecture best practices. There are well-known practices that help a software architect to understand a software architecture. Using our AI Accelerators powered by LLMs, it streamlines key stages of code modernisation, improving understanding, generating accurate documentation, and supporting better technical decision-making. 

Can you share a feature you are especially proud of? 

Valentin: A key feature of the platform is the ability to generate multiple diagrams that simplify complex information.  

Clément: This is accomplished using the PlantUML library, allowing the LLM to generate diagrams based on the user’s requirements. Diagram creation is fully flexible: every element can be edited. The LLM and the user collaborate through an interactive conversation to refine and optimise the diagram. Once the user approves the final version, it is saved and incorporated into the documentation. 

How do you approach personalisation for each client?  

Arthur: For this, we will work closely with the client to establish a modus operandi that can be replicated for future projects. At this stage, personalisation is primarily reflected in the document template provided by the customer. 

Valentin: By using document templates validated by the client, that closely reflect the ones they already use, we ensure clients can maintain their established practices. 

Clément: For generation in Microsoft Word, we start with the client’s Word template, keeping the main elements such as titles, subtitles, font sizes, and colours. We then insert markdown placeholders into the file: these placeholders act as variables that are replaced with the information captured during the conversation between the LLM and the user.  

How do you ensure the tool is easy to adopt for users?      

Valentin: By designing it to resemble a modern chatbot, we aim to create an experience that feels like an exchange between two coworkers in a text conversation. 

Clément:  The main difference lies in the project creation, which involves multiple steps and conversations rather than just separate discussions. We also developed a ‘Help Page’ that provides an introduction to the tool along with a quick tutorial. 

Arthur: The best way to make any tool easy to adopt is through its user interface and user experience. This is the crux of the matter for all companies: every user who feels there are too many interactions, that something is unclear, or that the effects of mouse interactions are not intuitive is a user likely to abandon the tool for another. It’s also my responsibility to anticipate these cases and share them with the team.