How PSI Software – Logistics Built a Flexible Cloud Platform for Warehouse Intelligence
PSI Software – Logistics is part of the PSI AG Group headquartered in Berlin, which has been developing software for industry for over 50 years. The Poznań branch of PSI has achieved success in implementing and developing the PSIwms system for warehouse management.
As part of the PSIwms system development, the team developed the Warehouse Intelligence concept, which involves using artificial intelligence to optimize logistics processes.
Services Used:
Client:
PSI Software – Logistics
Industry:
Manufacturing and Industry
SaaS / ISV
Technologies:
AWS
Kubernetes (EKS)
Terraform
GitLab CI
ArgoCD
ElasticSearch
LogStash
Kibana
Business Challenge
The PSI Polska team tasked us with designing an environment that would support the pace of research and development work on Warehouse Intelligence. We identified key technological barriers that needed to be resolved for the platform to become a real tool for AI model experiments, not just infrastructure support.
Need for rapid deployment of test environments:
Each new Machine Learning experiment required a separate, fully reproducible configuration. The time required to create environments and conduct tests became one of the main constraints on product development pace, necessitating a solution that would eliminate repetitive manual work.
Intuitive operation without deep DevOps knowledge:
The team developing Warehouse Intelligence focused on model logic and the PSI software itself, not on infrastructure management. The environment had to be accessible enough for individuals without advanced technical knowledge to independently manage all its components.
Experiments on different hardware configurations:
Machine Learning models required both standard CPU instances and machines with graphics cards. The platform had to support rapid switching between configurations and launching GPU resources exactly when needed—without excessive costs during periods of low utilization.
Risks of investing in proprietary hardware:
Purchasing proprietary infrastructure for an R&D project would mean the risk of incorrect hardware selection, rapid obsolescence, and scalability limitations. The client needed a model where computing resources appear and disappear in rhythm with projects, not capital investments.
Machine learning experiments that were previously blocked by hardware selection and environment configuration time are now conducted in parallel across multiple variants. We have received a platform that grows and shrinks in rhythm with our ideas, not with the procurement plan. This is a completely different level of comfort when working on a research product.
WMS Product Owner
Jerzy Danisz
PSI Software – Logistics
Cloud native platform designed for ML and GitOps
In response to PSI Polska’s needs, we built a fully cloud-based environment on AWS, entirely managed as code and prepared for cloud native operations. Our goal was to combine research flexibility with operational discipline so that the client’s team could focus on Machine Learning experiments themselves, not on server maintenance.
Scope of Work
Architecture Design and Infrastructure as Code
We began work with consultations and precise environment planning for testing on different virtual machine sizes. We described the entire infrastructure in Terraform, enabling PSI to launch new environment copies in multiple configurations simultaneously, with full versioning in GitLab and without the risk of manual errors.
Containerization and Kubernetes Orchestration
We built the environment on Amazon EC2 and Amazon EKS, which took over container orchestration. We placed all application components in containers, making the platform resilient to overloads, with the number of instances dynamically growing and shrinking with computational power demand and immediate launching of EC2 instances with GPU for ML testing when needed.
CI/CD Automation in GitOps Model
For the project, we launched GitLab and described all processes in CI/CD pipelines. We implemented ArgoCD in the Kubernetes cluster as a GitOps tool, making the code repository the single, consistent interface for managing the entire platform with full change auditing and deployment repeatability.
Monitoring, Data, and Traffic Management
Logs from experiments flow to the ELK stack (ElasticSearch, LogStash, Kibana), which we implemented for real-time monitoring and rapid problem resolution. We store output data in Amazon S3, and incoming traffic to the cluster is managed by a Network Load Balancer deliberately selected to keep the entire platform portable between public and private clouds.
R&D Environment That Keeps Pace with AI Ambitions
The implemented solution transformed infrastructure from a development barrier into a catalyst for the PSI Polska team’s research work. The Warehouse Intelligence platform gained repeatability, flexibility, and constant visibility, which directly translated into the pace of Machine Learning experiments and the comfort of those conducting them.
Key Results:
deployment speed
Time to launch a complete R&D environment using Terraform/IaC
GPU on-demand
Startup time for EC2 instances with GPU on-demand for ML testing
Operational Independence and Support Exactly When Needed
After platform deployment, the PSI Polska team independently manages the Warehouse Intelligence environment and develops it in rhythm with their own research work. Our role is to remain a technology partner available on an ad-hoc basis when the need arises for another improvement, consultation, or rapid response to a new business requirement.
Independent PSI Team in Daily Platform Operations
Thanks to foundations such as Infrastructure as Code, GitOps, and containerization, the PSI Polska technology team conducts daily operations and further environment development without the need for constant support from our side. All processes are described in code, fully repeatable, and transparent to the client’s engineers, which was one of the project’s objectives from the beginning.
