Dassault DSL Assistant
AI agent that generates scripts in a proprietary scripting language used by clients of a major industrial software vendor. A task once reserved for experts, done in minutes instead of weeks.
Context
Dassault Systèmes builds industrial software platforms used by thousands of companies worldwide — CATIA for 3D design, ENOVIA for product lifecycle management. Within these platforms, clients can write scripts in a proprietary programming language to automate business processes: automatically generating bill-of-materials, validating product configurations, triggering custom workflows. These scripts are a powerful tool — but until now reserved for specialists, after weeks of training.
Problem
This scripting language is entirely proprietary — no language model has ever seen it during training. And its documentation didn't exist in any usable form: scattered across internal XML files, database dumps, and heterogeneous sources in the internal datalake, never consolidated in one place. For an experienced developer, learning the language and writing a first useful script took several weeks: finding the right docs, understanding the syntax, testing, fixing. For non-technical client teams, it was simply out of reach.
Approach
Before touching any AI, the documentation had to be rebuilt from scratch. A complete ETL pipeline extracts, normalizes, and merges heterogeneous sources (XML files, database dumps, regex parsing) to reconstruct coherent documentation. This documentation is then structured into an RDF knowledge graph — a formal knowledge representation that lets the RAG engine reason about relationships between concepts, not just retrieve text passages. On top of this, a LangGraph agent orchestrates generation: the user describes their need in plain language, the agent retrieves relevant documentation, generates the script, automatically checks syntax, and corrects if needed. One hard constraint throughout: everything runs on lightweight open-source MoE models, no fine-tuning, no proprietary data sent outside — a non-negotiable requirement in an industrial R&D context.
Impact
What previously took several weeks — learning the language, training teams, writing the script — now takes minutes. A non-technical user describes their need in plain language and gets back a working script, ready for review. Clients can now deploy autonomously, with no prior training or dedicated support. Estimated 80% reduction in scripting time.