🦒 Girafon

Automatically analyze ESG/sustainability reports against ESRS (CSRD) requirements and surface what's missing with evidence.

#sustainability#reporting#eu#omnibus#esg#esrs#llm#greenwashing#csrd#efrag#ghg-protocol
View Demo View Code

Girafon main page

Overview

ESG teams face long, uneven reports and must assess ESRS coverage without a consistent structure.

Problem

Manual ESRS checks are slow and inconsistent, which increases compliance risk and review costs.

Target Users

  • ESG reporting teams
  • Sustainability consultants
  • Compliance and audit stakeholders

Goals

  • Handle heterogeneous PDFs without manual clean-up
  • Surface evidence with each gap to earn trust in LLM output
  • Keep reviews consistent across reports

Non-Goals

  • Final legal sign-off
  • Full report authoring
  • Long-term document storage

Solution

A report analysis flow that extracts PDF content, maps it to ESRS disclosures, and highlights missing or weak coverage with evidence.

My Role

Personal case study. I defined the evaluation logic, designed the analysis workflow, and built the demo end to end, including a hard choice to require quoted evidence for each finding, which reduced recall but improved trust.

Key Decisions & Trade-offs

  • Treated ESRS as a structured checklist to make gaps explicit and actionable.
  • Surfaced evidence alongside findings to build trust in the analysis.
  • Focused on the core flow over full report authoring features.

Key Features

  • ESRS coverage mapping with per-disclosure status
  • Gap list with evidence references and a prioritized fix list
  • Clear summary with a weighted score and greenwashing signals

AI Use

  • LLM used to extract and map report text to ESRS requirements.
  • Findings include quoted evidence and are reviewed as advisory outputs.

Risks & Constraints

  • Inputs vary widely in structure and quality
  • Models can misinterpret ambiguous language
  • Constraint: Most reports are heterogeneous PDFs, so I prioritized extraction reliability over broad coverage.

What I’d Do Next

  • If I restarted, I would focus on a smaller ESRS subset to harden evidence mapping before scaling.
  • Add a coverage score and heatmap by ESRS section.
  • Support multilingual reports and mixed-format PDFs.
  • Provide an audit trail for review and sign-off workflows.

Status / Deployment

Showcase demo. Demonstrates what is possible after self-deployment.

References