FinScribe

Overview

FinScribe is an AI-driven tool designed to automate financial document analysis for credit analysts. It processes PDFs such as annual reports and financial statements, extracts key data, and generates structured summaries with risk insights.

The system reduces manual effort, improves consistency, and enhances decision-making by combining document parsing, AI analysis, and validation workflows.

Key Features

• Automated PDF analysis with table and text extraction
• AI-generated executive summaries with confidence scores
• Key financial metrics detection with trend indicators
• Risk identification with severity levels
• Source tracing that links insights back to PDF pages
• Document Q&A chat interface
• Secondary validation using another AI model
• Export reports in Markdown/Text format

How It Works

Upload PDF -> Extract Data -> AI Analysis -> Validation -> Final Memo -> Export

• Input: Financial PDF document
• Processing: Extract tables and text, then analyze with AI
• Validation: Secondary AI checks accuracy
• Output: Structured credit memo draft

Tech Stack

• Frontend: Streamlit
• Backend: Python
• PDF Processing: pdfplumber
• Data Handling: pandas, numpy
• AI / LLM: OpenAI (GPT-4o, GPT-4o-mini)
• Environment: python-dotenv

Output

The system generates:

• Executive summary with bullet points
• Key metrics with trends
• Top risks with severity levels
• Confidence scores for each insight

Example: strong data and incomplete data flags for quick review.

Advanced Features

• Pseudonymization for data privacy before sending to LLM
• Token usage and cost monitoring
• Feedback-based memo regeneration
• Support for password-protected PDFs
• Optional local LLM support (Ollama)

Use Cases

• Credit analysts preparing reports
• Financial institutions analyzing documents
• Students learning financial analysis
• Automating due diligence workflows