Recomentor System Architecture
30/05/2025
Recomentor is a SaaS platform integrating user authentication via Keycloak, document annotation and metadata management by PoolParty, ESG indicator processing through the ESG API, and semantic storage using GraphDB and the Elasticsearch index. It uses OpenAI’s API for generating report summaries and utilizes Dagster pipelines for automated data collection and processing.
Recomentor is a Software as a Service (SaaS) application with the following system architecture:
 |
User authentication on Recomentor is managed by Keycloak, which is connected to PoolParty, the ESG API, the UI, and the Java back end. Keycloak provides separate realms for PoolParty and Recomentor users. For more information on how Recomentor keeps your data safe, visit our Secure Authentication & Data Storage help page.
The ESG-API manages the following:
Requesting indicator scores for compliance checks
Uploading documents
Summaries generated by the LLM
Paragraph retrieval
The Java back end uses GraphDB to store information on companies, frameworks, and peer filters. The ESG API retrieves documents via elasticsearch for indicator calculations. PoolParty is used to upload and annotate documents and to retrieve information about the ESG taxonomy. The OpenAI API summarizes paragraphs whereas GraphDB supports both metadata and report queries.
Important
An LLM call is sent to the OpenAI API whenever a user requests a generated report summary by clicking on a category from the General Compliance area. Fragments of reports relating to the ESG concept of interest are used as input for this prompt.
The ESG API also uses the Dagster data pipeline service for web scraping, which includes the collection of documents as well as the collection and processing of company data.
Report files, such as PDFs, are stored in and linked from an AWS S3 bucket.