Shippeo, a leading platform for supply chain visibility, is taking a new step forward with the launch of Shippeo AI: an embedded intelligence layer that transforms real-time data into concrete actions.
Testimonial Shippeo AI - Turning Supply Chain Visibility into Action
Shippeo is a company dedicated to supply chains that provides real-time tracking data on all transport processes.
To move from a simple information system to a real engagement system, Shippeo launches Shippeo AI, a library of AI agents and calls on Theodo to develop the first two use cases:
- one Talk-to-My-Data Agent allowing users to better understand possible quality problems with Shippeo data (incomplete tracking...)
- a second operational officer to answer any question about a transport order.
The challenges to be met:
- Create agents in a complex and microservices system with multiple data sources (Data Warehouse, API, Algolia) ready for production with a custom front end interface
- Create a low-latency Data Analysis Agent capable of understanding business data quality issues.
- Orchestrating multiple agents to answer user questions
Our approach
What made the difference:
An agent architecture designed to adapt to all situations
For an AI agent to be really useful in production, it must be able to manage very varied requests, from the simplest to the most complex. We have adopted a “Workflows Backed Agents” approach that combines the flexibility of AI with the reliability of structured processes. Result: the agent knows when to improvise and when to follow a marked path, guaranteeing performance and relevance regardless of the request.
.png)
Tools to assess and improve the quality of responses
An AI agent is only valuable if its answers are reliable. To ensure this, we have developed an evaluation interface with Streamlit, which allows teams to easily check the queries generated by the agent and keep the reference data sets up to date. This device makes it possible to quickly identify errors, to iterate and to improve the accuracy of the agent on an ongoing basis.
.png)
A product designed to last, technically and humanly
Delivering a successful agent is not enough if the customer cannot make him live on time. We have implemented monitoring tools such as Langfuse and rigorous development standards to guarantee the resumption of the project independently. At the same time, we worked on the user experience in depth: the agent understands the context of the page on which the user is located and does not just respond, he proposes adapted follow-up actions, transforming each interaction into real support.
.png)
We were not interested in delivering a shiny POC. We were relentless in delivering customer value in a complex production setting.
