Skip to main content

brookfields_engg

2. PEP (Politically Exposed Person) Checks​

Inputs & Current Setup:

  • Data Source: Using Acuris for PEP checks.
  • Current Environment: Checks are currently being executed using the development (dev) API endpoint.
  • Cost Consideration: If we switch from the dev API to the production API, each PEP check costs approximately INR 106 (subject to change).

Currently Available:

  • PEP Level Identification:
    • We have direct access to a PEP dataset through Acuris.
    • The system can identify PEP levels (e.g., Tier 1, Tier 2, Tier 3 exposure).
  • Dev API Integration:
    • The integration with the Acuris dev environment is functioning and allows us to perform PEP checks.

Required Work:

  1. Production API Integration:
    • Move from the Acuris dev environment to the production API endpoint.
    • Validate authentication, request quotas, and implement cost-tracking mechanisms to handle INR 106 per check.
  2. Refine Matching Logic for Connected Parties:
    • Enhance the entity resolution to apply PEP checks not only to direct Targets and Directors but also to all connected parties (e.g., shareholders, beneficiaries).
    • Integrate date-of-birth, nationality, and other attributes to reduce false positives.
  3. Risk Categorization & Reporting:
    • Ensure that PEP risk levels (Tier 1, 2, 3) are clearly integrated into the final reports.
    • Include logic to automatically flag high-risk PEPs and generate corresponding alerts or notifications.

4. Adverse Media and Non-Adverse Media Analysis​

Inputs & Current Setup:

  • Classification Columns: A predefined set of columns represent various events and signals (e.g., acquisition-acquiree, legal-issue, regulatory-issue, user-growth).
  • Planned Expansion: We need additional signals, primarily related to adverse media. Current classification setup includes both adverse and non-adverse categories, but we need to add more signals beyond the 40+ currently listed.

Currently Available:

  • Base Classification Framework:
    • A preliminary classification schema with columns like legal-issue, regulatory-issue, financial-challenge, and more.
    • This framework can tag media articles based on keyword or pattern matches, aiding in initial filtering.

Required Work:

  1. Additional Signal Integration:
    • Incorporate new adverse media signals into the classification model.
    • Update mapping logic to categorize media hits under newly added signals related to adverse events (e.g., bribery, corruption, fraud).
  2. ML/NLP Model Improvements:
    • Move from simple keyword-based rules to more advanced NLP techniques (e.g., transformer-based models) to improve accuracy of classification.
    • Fine-tune the model using historical datasets to reduce false positives and negatives.
  3. Semantic Search & Vector Database Integration:
    • Fully integrate the vector database for semantic similarity searches, enabling quick retrieval of relevant media articles by topic.
    • Ensure that the database can handle newly added signals efficiently.
  4. Summarization & Timeline Visualization:
    • Implement summarization algorithms to provide concise overviews of adverse media over time.
    • Add timeline views to show when certain signals (e.g., "legal-issue", "regulatory-issue") appeared, providing historical context and trends.