Serverless AI-Enhanced ESG Platform

Technical Specification: Serverless AI-Enhanced ESG Platform

1. Project Overview

A multi-tenant, serverless SaaS platform for ESG (Environmental, Social, and Governance) reporting and verification. The platform allows companies to upload sustainability guides and data, which an AI then analyzes to provide verified commentary for bankers and regulators.

2. Core Pillars (Non-Negotiable)

Pillar 1: AI RAG Engine (Verification Logic)

Dynamic Context Switching: The AI must switch context based on the selected Guide (e.g., NoS-F, GRI 11). Logic must use metadata filtering (e.g., guide_id) to prevent "sticky" context.

Evidence Vault (Citations): Every AI-generated claim must include a source citation (Document Name + Page Number).

LLM: AWS Bedrock (Claude 3.5 Sonnet).

Pillar 2: Cost-Efficient Architecture

Vector Database: PGVector on Amazon Aurora Serverless v2.

Constraint: Do NOT use Amazon OpenSearch or provisioned services. The goal is $0/month idle cost (true serverless).

Compute: AWS Lambda or AWS App Runner (Auto-scaling).

Pillar 3: IoT & MRV Ingestion

API Endpoint: A secure POST endpoint for JSON telemetry (DeviceID, Timestamp, Metric_Value).

Tenant Guard: Data must be isolated at the database level using a Tenant ID middleware. No user should ever see another company's data.

Pillar 4: Multi-Tenant Dashboard

Frontend: React/Tailwind.

Features: PDF Uploader, Multi-Guide Selector, IoT Analytics Chart, and AI Commentary Panel with Clickable Citations.

3. Milestone 1 Deliverables (The "Bank-Ready" MVP)

Successful ingestion of PDF guides into PGVector.

AI Chat/Analysis that filters by Guide ID and provides page-level citations.

Functioning IoT API endpoint with data reflected on a tenant-specific dashboard.

Infrastructure deployed via Terraform or AWS CDK (Serverless only) Show More