Background:
A global bank is the largest source of multilateral development financing for Latin America and the Caribbean, supporting economic development, social development, and regional integration by lending to governments and government agencies, including State corporations. The Bank is owned by 48 sovereign states, which are its shareholders and members. It provides loans, grants, guarantees, policy advice, and technical assistance to the public and private sectors of its 26 borrowing member countries.
Challenge:
The Bank has gathered structured and unstructured documentation regarding its various initiatives and projects over many years. Valuable insights and observations have been captured within this vast body of documentation around the context of “Lessons Learned”. They needed a solution that would quickly search tens of thousands of documents and surface the “Lessons Learned” to help ensure future success. Critically, a multi-lingual solution was required, as the Bank operates in four languages (English, French, Portuguese, and Spanish).
Converge was engaged to migrate the Bank’s legacy application and build an internal search engine specific to their needs. An outdated legacy system written in J2EE on IBM WebSphere Application Server was found to be unwieldy and lack the agility to scale to the processing and data analytic demands of new projects and ventures.
The Bank had previously successfully partnered with Converge on a cloud migration project for an AI-driven advanced search application rewritten for cloud deployment, to help locate Subject Matter Expert staff members with tacit knowledge within the Bank. Deep Machine Learning, Natural Language Processing, and additional artificial intelligence techniques drive the underlying document parsing, indexing, and ranking. These AI technologies create an autonomous pipeline for intuitive, accurate, context-aware searches. Converge was selected to repurpose these AWS-based AI capabilities and user experience toward the migration and modernization effort of the development of “Lessons Learned”.
Solution:
The application was migrated from the Bank’s existing on-premise IBM WebSphere Application Server to AWS. Converge developed a system which provides a cross-language search experience that identifies key insights and knowledge (“Lessons Learned”) from prior financing projects. The team leveraged several neural network models to provide query expansion and semantic search across the four different languages. Customized word2vec word embeddings and transformer models are the AI/ML engine of this solution.
The four-language “Lessons Learned” search engine provides contextual, intelligent, and accurate searches using and training AI & machine learning models. It includes an administrative dashboard with “likes”, general activity metrics, and case definition for integration with other services.
The system delivers advanced AI/ML features for Boolean Searches, prioritized search algorithms, search by objectives, similar lessons, and semantic queries, by leveraging many AWS services, including core AWS services of API Gateway, Elasticsearch, and SageMaker.
Results:
The Converge team delivered a revolutionary user experience in search technology for the Bank. With one click, users can identify similar and related lessons across four languages, with a search powered by cutting-edge deep learning models.
The fully functional API also serves content across numerous services at the Bank. This AWS-native application uses infrastructure-as-code for DevOps, is fully integrated with bank ETL pipelines, and is automatically updated using state-of-the-art DataOps and ModelOps processes.
The Bank has high hopes this custom trainable Lessons Learned quad-lingual search engine can be expanded to serve multiple AI/ML models of a host of applications across the Bank.
