AI-Powered ESG Scoring Platform

Problem

ESG evaluation for 2,000+ listed companies was a manual, analyst-heavy process spread across thousands of PDFs, XBRL filings, and annual reports — slow to scale and hard to audit.

Approach

• ▸Designed an end-to-end RAG pipeline: vector embeddings over ingested filings, combined with live web grounding via Tavily search and Gemini Search Grounding for facts outside the document set.
• ▸Shipped the scoring service as Kubernetes-deployed FastAPI microservices, with a Generated → In Review → Finalized workflow and full audit history for analyst sign-off.
• ▸Built the ingestion layer in Airflow to scrape and process 2,300+ ESG PDFs, XBRL filings, and annual reports across multiple financial years, with idempotent processing.
• ▸Modeled the data in PostgreSQL — schemas, audit tables, and materialized views — to back a real-time market dashboard with 1-day to 10-year historical charting.

Impact

• ✓Automated ESG evaluation for 2,000+ listed companies, cutting manual analyst effort.
• ✓2,300+ source documents processed with 99% pipeline reliability.
• ✓Production service running on Kubernetes with a full audit trail from generation to publish.