Talk
Tracking products across a supply chain is challenging due to fragmented, semi-structured, and often inconsistent data sources. This talk presents a production-oriented system that combines NLP and graph-based modeling to reconstruct product lifecycles and identify anomalies.
The system consists of three main components:
• NLP Layer: Processes receipts, invoices, and customs declarations to extract structured attributes
such as product category, brand, and unit measures from noisy text data.
• Graph Layer: Links suppliers, buyers, and transactions into a network, enabling end-to-end tracing
of goods across multiple stages of the supply chain.
• Analytical Layer: Compares declared and observed values (prices, quantities, volumes) to detect
inconsistencies and potential fraud patterns.
The talk focuses on practical challenges such as handling multilingual and inconsistent text, designing scalable pipelines, and combining semantic and structural signals effectively.
Key takeaways:
• Designing robust NLP pipelines for semi-structured transactional data
• Using graph models for traceability and relationship analysis
• Detecting anomalies through combined text and graph signals
• Trade-offs between scalability, accuracy, and system complexity
Audience: Data scientists, ML/AI engineers. Basic familiarity with machine learning is sufficient.
About the Speaker
Naira Barseghyan is an ML/AI Engineer specializing in applied NLP and large-scale data systems. She works at Mindwise LLC on a World Bank–funded government modernization program, developing solutions that improve transparency and analysis at a national level.
She has experience improving system performance at scale, reducing processing time for millions of records, and building machine learning systems that operate reliably in real-world environments. Her work focuses on delivering practical solutions that turn complex data into meaningful insights and support better decision-making.
Naira holds a BSc in Data Science from the American University of Armenia. She is particularly interested in applying machine learning to real-world challenges and contributing to systems that have tangible impact.