LCM Go Cloud helps Cosmos Aluminium accelerate information retrieval

At Cosmos Aluminium, the need to improve operational efficiency was most apparent in two departments: accounting and HR. The HR team often managed more than 1,000 résumés, each formatted differently and using inconsistent terminology. Finding suitable candidates for an open position required manually reviewing documents, a process that took an average of 14 days.

In addition, the company had produced 170 manuals to standardize processes in SAP enterprise resource planning (ERP). Employees who needed guidance on those processes manually searched documents, consulted colleagues, or submitted help desk tickets, often waiting up to 30 minutes for answers. This approach slowed decision-making and introduced inconsistencies across workflows. “The main scope of this project is to search quickly,” says Petros Chondronasios, IT manager at Cosmos Aluminium. “The process was manual and slow, relying heavily on individual experience. This often caused delays and inconsistent results.”

Building on an existing relationship, Cosmos Aluminium selected LCM Go Cloud because of the partner’s experience in delivering generative AI solutions on AWS. The solution consists of two applications: The CV Intelligence Application evaluates résumés against a job’s requirements. The Accounting Manual Application searches SAP ERP manuals for answers to users’ questions. To create the generative AI agents, the team used Amazon Bedrock, a service for building generative AI applications and agents at production scale. The agents rely on Amazon Bedrock Guardrails to implement safeguards and Amazon Bedrock Knowledge Bases to deliver more relevant, accurate, and customized responses. To parse content and identify similar skills that are described differently, the CV Intelligence Application uses the Claude Sonnet and Haiku foundation models in Amazon Bedrock.

Running 14 functions across six queues, the architecture processes documents asynchronously by using AWS Lambda, which handles all infrastructure management. The solution stores raw and processed documents across five buckets in Amazon Simple Storage Service (Amazon S3), which offers industry-leading scalability, data availability, security, and performance. To keep metadata and session history, the solution uses Amazon DynamoDB, a serverless, fully managed, distributed NoSQL database with single-digit millisecond performance at virtually any scale. The team implemented secure, scalable authentication and access control through Amazon Cognito so that users can access only the information relevant to their responsibilities.

To connect on-premises systems to AWS resources, the team chose AWS Site-to-Site VPN, which establishes secure and private sessions. Security controls include encryption at rest and in transit, antivirus scanning of uploaded files, and restricted access to keep data within the European Union. By using Amazon CloudWatch, the manufacturer can observe and optimize its workloads at nearly any scale. To help users interact with data in their preferred language, the solution provides a bilingual chat interface in Greek and English. Users can request information in natural language and receive responses that are grounded in business documents. The interface also supports document management so that users can upload, search, and retrieve résumés and ERP manuals within a single system.

The solution significantly reduced the time required to retrieve information across accounting and HR departments. The Accounting Manual Application now returns near-instant responses from 170 ERP manuals, replacing workflows that previously required up to 30 minutes. The CV Intelligence Application cut candidate search time from 14 days to approximately 15 minutes to find and compare candidates, helping HR teams identify and compare candidates as a job becomes available. Users of both systems required minimal onboarding and no formal training beyond a short orientation session. “Built on AWS, the solution is simple to use, requires little to no training, and was quickly adopted by 25 users from day one,” says Chondronasios.

The project moved from kickoff to production in 16 weeks. And the solution exceeded its 95 percent accuracy baseline, with estimates closer to 99 percent in practice. These improvements streamlined internal processes, reduced dependency on manual expertise, and increased consistency across both workflows.