Project Thoth is a recommendation engine that collects information about software packages, container images such as installation, assembling issues, runtime crashes or information about performance. This information is subsequently used in a recommendation engine that searches large state space of libraries and recommends the best possible combination of libraries suitable for your application using reinforcement learning. Let’s have a look at how such information is collected and how the large state space is explored to resolve the best application stack for your Python application based on different aspects. Python ecosystem is experiencing significant growth and popularity especially with the hype machine learning, data science and AI are creating. As the ecosystem grows its many times not straightforward and easy to decide which libraries in which versions are the most suitable ones for an application. Project Thoth is a recommendation engine which aggregates various characteristics of Python packages, called "observations", and uses them to recommend the best possible software stack (a fully pinned down list of dependencies) suitable for user's runtime environment and the application purpose. In this talk, we give an overview of the project Thoth, main ideas in data aggregation and its recommendation engine based on reinforcement learning principles. We will also show how you can benefit from Thoth's recommendations.