About Lokad
Paris-based supply chain software company whose platform runs forecasting and optimization on its own scripting language, Envision.
Lokad builds supply chain optimization software. Its clients describe forecasting and inventory logic in Envision, a domain-specific language created by Lokad. Envision is powerful, but new users need time and help to write their first scripts.
OSKI had already worked with Lokad on a CRM and UX project between 2023 and 2024. In 2025 the team was asked to put an assistant directly into the platform.
What had to be solved
The assistant had to do two different jobs in one chat window and do both reliably.
- Answer general questions about the platform and its modules, and suggest which Lokad specialist to contact.
- Recognize when a user needs code, switch into an Envision mode and generate scripts that actually run.
- Remember each user's conversation across messages in a real-time, event-driven system.
- Render Envision code readably inside chat messages and push notifications in real time.
How the team built it
Before writing prompts, the team studied how people used the platform and Envision: the questions they asked most and where they got stuck. Those patterns became the assistant's core behaviors.
A multi-step check decides whether a message calls for an explanation or for code. In Envision mode the assistant writes the script and formats it with proper highlighting in the chat.
The assistant connects securely to the Envision runtime and runs generated code there. Scripts that fail go back for another pass.
Conversation memory sits on Lokad's own Azure event store, so context survives across messages. When a question needs a person, the assistant suggests the right Lokad contact.
What changed
The assistant shortened the path from a question to a working script and reduced time lost looking for help.
- Users reach a first working Envision script about 70% faster.
- 82% of generated scripts run on the first attempt and 94% by the second.
- Relevant team contacts are suggested in about 78% of cases.
- Time spent searching documentation or the right person dropped by about 60%.
“This is a team of engineers who obviously like what they are doing.”
Inside the product
