Helping DG EAC Prioritise Similar Erasmus+ Applications for Review
The Challenge
DG EAC faced a growing operational challenge as Erasmus+ generated thousands of applications that needed to be reviewed fairly, consistently and at scale. For National Agencies, identifying potentially plagiarised or overly similar submissions was especially difficult because manual review across large volumes of narrative content is slow, subjective and resource-intensive. Reviewers needed more than a simple text-matching tool: they required a secure, evidence-based platform that could highlight meaningful similarities, surface copied wording and support defensible decision-making. The complexity was increased by the structure of Erasmus+ KA122 applications, where similarities may appear in specific sections rather than across entire documents, and where legitimate overlap must be distinguished from suspicious duplication. DG EAC therefore needed a solution that could process application content securely, compare submissions intelligently and help reviewers focus on the cases most likely to warrant investigation. The goal was not to automate final judgments, but to equip National Agencies with a practical review workflow that reduced noise, improved consistency and strengthened anti-plagiarism efforts within a sensitive public-sector context.
Our Solution
Nesqual Tech designed and delivered a secure similarity review platform tailored to the Erasmus+ KA122 assessment process. The solution combined semantic analysis with shared-wording detection to flag applications that were potentially too similar, then presented the evidence in a reviewer-friendly interface so National Agencies could make informed decisions. Rather than relying on keyword matching alone, the platform compared same-section texts to identify both meaning-based similarity and direct overlap in wording, improving the relevance of the cases surfaced for review.
The application was built as a secure Azure-based web platform, with a modern interface for case review and a Databricks-powered backend for large-scale evidence processing. Reviewers could access ranked cases, inspect side-by-side comparisons, view copied-passage highlights and use English translations where needed. Decision history and export capabilities ensured that findings could be recorded, shared and revisited, supporting transparency and auditability. The architecture used Azure Container Apps, Azure Application Gateway WAF and Microsoft Entra ID to provide secure access and deployment, while Azure Databricks, Apache Spark, Delta Lake and Unity Catalog supported scalable data processing and governance. AI capabilities were integrated through Azure AI Foundry, OpenAI text-embedding-3-small, Databricks Model Serving and Anthropic Claude to help detect nuanced similarity and support evidence interpretation. Delivered over six months by a team of four, the platform gave DG EAC a focused, technically robust pilot for anti-plagiarism review without removing human oversight from the final decision process.
Key Results
- Analysed a pilot set of 50 Erasmus+ KA122 applications, covering 885 section texts and 20,638 same-section comparisons.
- Identified 729 application pairs above the similarity threshold, then prioritised 10 cases for first review—just 1.4% of flagged pairs—while keeping the remaining 719 available for follow-up.
- Found that 5 of the 10 priority pairs shared wording across five sections of the same two applications, while the other 5 showed unusually high section-level similarity that warranted closer investigation.
Technologies Used
Have a project in mind?
Get an instant AI price estimate for it, or talk directly to our team.
One email a month on what we learn building with AI