TriSE

Siemens faced the critical challenge of navigating and making sense of vast, fragmented, and continuously evolving technical knowledge spread across heterogeneous public sources such as arXiv, GitHub, and StackOverflow. Traditional search and analysis tools were insufficient for identifying emerging trends, synthesizing cross-source insights, or supporting domain experts in timely, data-driven decision-making.

fortiss designed and implemented a platform which integrates multiple public data sources (arXiv, GitHub, StackOverflow, etc.) and applies machine learning for trend detection, knowledge graph construction, and intelligent search. The platform includes three main modules—Relations, Statistics, and Q&A—built on technologies like Word2Vec, BERT, Neo4j, and Elasticsearch. fortiss also created a user interface that allows Siemens experts to define, explore, and refine topic areas through a semi-automated process supported by generative AI. Iterative work packages aligned with Siemens’ needs enabled the platform to evolve with real-time data pipelines, LLM-based domain knowledge expansion, and hands-on analytics.

• Automated Data Integration & Trend Updates: Built a robust ETL pipeline with Apache Airflow to continuously ingest and refresh diverse software engineering data, ensuring up-to-date and relevant trend insights.
• AI-Driven Knowledge Expansion: Developed a semi-automated domain knowledge builder using large language models (LLMs) to broaden trend coverage and accelerate topic discovery.
• Expert-Centric Exploration Tools: Delivered an intuitive UI with intelligent search, trend scoring, and customization features—enabling Siemens experts to define, refine, and visualize key technology areas.

01.02.2024 - 31.12.2024