README.md

metadata

title: ESG Document Intelligence Platform
emoji: 🌿
colorFrom: green
colorTo: blue
sdk: gradio
sdk_version: 6.9.0
app_file: app.py
pinned: false
license: apache-2.0
short_description: HyperRAG + Discourse Graph for ESG Report Analysis

🌿 Multimodal ESG Document Intelligence Platform

HyperRAG + Discourse Graph Reasoning for ESG Report Analysis

Upload any ESG / Sustainability PDF report and get:

• 💬 Contextual Q&A — ask questions about the report, answered with page-level evidence
• 📊 ESG Pillar Scores — keyword-based E, S, G scoring + sector detection
• 🚨 Greenwashing Detection — flags unsubstantiated claims with exact page references
• 🕸️ Discourse Graph Insights — models relationships between claims, evidence, policies and metrics

Architecture

PDF → Text Extraction (pdfplumber)
    → Chunking (400-word windows, 80-word overlap)
    → Embeddings (sentence-transformers/all-MiniLM-L6-v2)
    → Qdrant Vector Index (in-memory)
    → Discourse Graph (NetworkX DiGraph)
         claims ──supported_by──▶ evidence
         policies ──measured_by──▶ metrics
    → HyperRAG Retrieval
         vector search + graph neighbourhood expansion
    → Flan-T5 Answer Generation

Key Technologies

Layer	Technology
Vector Store	Qdrant (in-memory)
Embeddings	all-MiniLM-L6-v2
LLM	google/flan-t5-base
Graph	NetworkX DiGraph
Retrieval	HyperRAG (vector + graph)
UI	Gradio

Usage

1. Upload an ESG report PDF in the Upload & Process tab
2. Click Process Document — wait ~30–60 s for indexing
3. Switch to any analysis tab and explore!

Limitations

• ESG scores are keyword-density heuristics (not certified ratings)
• flan-t5-base is used for CPU compatibility; swap in a larger model for production
• Greenwashing detection is pattern-based and requires expert review

Running Locally

git clone https://huggingface.co/spaces/<your-username>/esg-intelligence
cd esg-intelligence
pip install -r requirements.txt
python app.py

License

Apache 2.0 — research & demonstration use only.