development and implementation of a search module into an ERP system deployed on the customer's side
The developers had two main tasks - to achieve a high level of relevance of search data and to create a system that can independently learn without operator intervention. Apache Lucene technologies were used to solve search problems, conceptual search methods - the method of dense vectors and neural networks. One of the varieties of recurrent neural networks, an LSTM network with fine-tuning capabilities, was used as neural networks.
Java, Python (Pandas, Keras), frameworks TensorFlow, Elasticsearch, Apache Spark
75% of the project was completed.