🌿 Eco-Truth: AI Greenwashing Auditor & Risk Sentinel The first autonomous AI Agent that forensically audits corporate sustainability reports against real-world evidence. --- ## 💡 What is this Actor? Eco-Truth Auditor is a specialized FinTech compliance tool designed to detect "Greenwashing" (misleading environmental claims) in corporate PDF reports. Unlike standard summarization tools, this Actor acts as an Adversarial Auditor. It does not trust the document. Instead, it extracts specific claims (e.g., "We reduced emissions by 50%") and aggressively cross-references them against external "Adverse Media" (news, lawsuits, fines, and NGO reports) to find contradictions. ### 🚀 Key Capabilities * Visual Forensics: Uses Multimodal AI to analyze charts and graphs for manipulation (e.g., truncated Y-axes, cherry-picked data ranges). * Adverse Media Screening: Automatically scrapes Google News for lawsuits, pollution scandals, and regulatory fines related to the company. * Claim Verification: Extracts hard claims from dense PDF tables and validates them against external ground truth. * Risk Scoring: Generates a normalized 0.0 - 1.0 Greenwashing Risk Score for instant credit/risk decision-making. --- ## 🛠️ How It Works (The Agentic Pipeline) This Actor orchestrates a complex, multi-step workflow: 1. Ingestion (LlamaParse): Downloads the target PDF and uses advanced computer vision to extract text, tables, and descriptions of charts. 2. Investigation (Deep Search): Triggers a secondary Actor to scrape the web for "fraud," "lawsuit," and "environmental violation" keywords linked to the company. 3. Forensic Analysis (GPT-4o): An LLM acts as a Senior Compliance Officer, comparing the internal claims (Step 1) against external reality (Step 2). 4. Reporting (Structured Data): Outputs a strict JSON report with a calculated risk score and specific evidence of contradictions. --- ## 📥 Input Parameters The Actor takes a simple JSON input specifying the target company and the document to audit. | Field | Type | Required | Description | Example | |---|---|---|---|---| | company_name | String | Yes | The name of the entity being audited. Used for adverse media search. | H&M Group | | target_url | String | Yes | Direct URL to the PDF Sustainability Report or ESG Disclosure. | https://hm.com/.../report-2024.pdf | Example Input JSON: ```json { "company_name": "The Coca-Cola Company", "target_url": "https://www.coca-colacompany.com/content/dam/company/us/en/reports/2023-environmental-update.pdf" } 📤 Output Data The Actor stores its result in the default dataset. The output is strictly typed and ready for integration into FinTech dashboards or Risk Management Systems. Key Output Fields: risk_score (Float): 0.0 (Safe) to 1.0 (High Risk). contradicted_claim (String): The specific claim from the PDF that was proven false. contradiction_detail (String): The evidence finding (e.g., "News report confirms a $5M fine for plastic pollution.") final_recommendation (String): Actionable advice (e.g., "Proceed with Caution"). Example Output JSON: JSON { "company_name": "The Coca-Cola Company", "risk_score": 0.45, "claims_found": 12, "contradictions_found": 1, "contradicted_claim": "We aim to collect and recycle a bottle or can for each one we sell by 2030.", "contradiction_detail": "Recent reports indicate virgin plastic usage has actually increased by 3% despite recycling claims.", "key_contradiction_source": "https://www.theguardian.com/environment/...", "final_recommendation": "Medium Risk - Verify plastic reduction targets manually." } 🎯 Use Cases: Green Finance & Lending: Banks can use the risk_score to automatically adjust interest rates for "Sustainability-Linked Loans" (SLLs). ESG Investing: Asset managers can screen portfolios for hidden compliance risks before they become public scandals. Supply Chain Diligence: Retailers can audit their suppliers' sustainability certificates for authenticity. Journalism & NGOs: Rapidly fact-check hundreds of corporate reports during annual disclosure cycles. 💰 Cost & Consumption: This Actor is optimized for high-value, low-volume runs. Compute: Uses ~1 minute of compute time per 50-page PDF. LLM Costs: Uses GPT-4o-mini for cost efficiency while maintaining reasoning quality. Search Costs: Performs 1 deep Google Search per run. ⚖️ Disclaimer This tool provides an AI-generated risk assessment based on available public data. It does not constitute legal or financial advice. All 'Risk Scores' should be verified by a human analyst before making financial decisions.