Agentic AI System for Climate Action Plan (CAP) Extraction & Benchmarking.
.........................................................................................................................................................................................................................
This project implements an agentic AI pipeline to extract, validate, and benchmark Climate Action Plans (CAPs) from cities and organizations. Unlike typical LLM workflows, this system is designed with.
- Explicit task decomposition (agents)
- Strict output structure (JSON schemas)
- Low-hallucination constraints
- Deterministic post-processing
The goal is to transform unstructured climate policy documents into auditable, comparable datasets for ESG analysis, climate strategy benchmarking, and decision support.
.........................................................................................................................................................................................................................
This system is built with a strict rule:
If information is not explicitly present in the CAP text, it must not be generated.
- Schema-constrained outputs
- Empty defaults instead of inferred content
- No speculative completion
- Explicit field-level extraction only
.........................................................................................................................................................................................................................
Missing data → "" or []
NOT → fabricated explanation
.........................................................................................................................................................................................................................
Raw CAP Text
→ Extraction Agent (LLM, schema-constrained)
→ Structured JSON (no inferred content)
→ Validation Layer (schema enforcement, null handling)
→ Aggregation (CSV dataset)
→ Deterministic Scoring Engine
→ Analytical Visualizations