AI Proposal Builder

This script automates the creation of auction proposals by combining information from source documents (text, PDF, images) with predefined templates using AI (OpenAI GPT-4o).

Features

• Extracts text from TXT files and PDFs (using OCR via Tesseract/Poppler as fallback).
• Optionally analyzes images in the source folder using GPT-4o multimodal capabilities to generate an inventory description.
• Uses predefined text templates (templates/ directory) for different proposal types (Personal Property, Real Estate, Combined).
• Calculates key dates (proposal date, acceptance deadline, ad start) automatically.
• Prompts user for auction duration to calculate end date before folder selection.
• Uses OpenAI GPT-4o for:
  • Analyzing the chosen template to identify variables.
  • Extracting information from source documents (including bios and optional photo description) to fill variables.
  • Generating the final proposal in Markdown format.
• Interactively prompts the user to fill in any information the AI couldn't find, including all variables marked as user in the template index (e.g., retainer, buyer's premium).
• Reports OpenAI token usage for each API call.
• Checks for required external dependencies (Tesseract, Poppler) on startup.

Variable Handling

• Variable Sources: Each template variable is marked as calculated, extracted, or user in the template index JSON.
  • calculated: Dates and totals computed by business logic.
  • extracted: AI attempts to extract from documents, user is prompted if missing.
  • user: Always prompted for user input (e.g., retainer, buyer's premium).
• Currency Formatting: All currency variables are formatted as plain numbers (no $), and the template handles currency symbols.

Setup

1. Clone Repository: Get the code onto your local machine.
2. Dependencies:
   • Python: Ensure you have Python 3 installed.
   • External Tools:
     • Tesseract OCR: Required for reading text from PDFs/images. Installation varies by OS:
       • macOS: brew install tesseract
       • Debian/Ubuntu: sudo apt-get install tesseract-ocr
     • Poppler: Required by pdf2image for PDF processing. Installation varies by OS:
       • macOS: brew install poppler
       • Debian/Ubuntu: sudo apt-get install poppler-utils
     • Ensure both tesseract and pdftoppm (from Poppler) are in your system's PATH.
   • Python Packages: Create and activate a virtual environment, then install required packages:

     python -m venv venv
     source venv/bin/activate # On Windows use `venv\Scripts\activate`
     pip install -r requirements.txt

3. API Key:
   • Create a file named .env in the project's root directory.
   • Add your OpenAI API key to the file like this:

     OPENAI_API_KEY=your_openai_api_key_here
     

4. Configuration:
   • Review and modify config.json if needed (e.g., change the default OpenAI model).
5. Templates:
   • Review and polish the .txt files in the templates/ directory. Ensure {{variable_names}} match expected data.
   • Review templates/will_mclemore_bio.txt and templates/mac_bio.txt.
6. Prompts:
   • Review the prompt files in the prompts/ directory. These control how the AI behaves.

Usage

1. Activate Environment: source venv/bin/activate
2. Prepare Data: Place your source documents (text files, PDFs containing details, relevant images) into a specific folder.
3. Run Script: Execute the main script from the project root:

   python main.py

4. Follow Prompts:
   • The script will first check for Tesseract/Poppler.
   • It will prompt you for the number of weeks until the auction before folder selection.
   • It will ask you to select the data folder using a file dialog.
   • It will ask you to choose the template type (1, 2, or 3).
   • If images are found in the data folder, it will ask if you want to generate a description from them (this uses the multimodal API and may take time/cost money).
   • It will process files and call the OpenAI API.
   • If information is missing, it will prompt you to enter values for all user and missing extracted variables.
   • Token usage for API calls will be printed.
5. Output: The final proposal will be saved as generated_proposal.md (or as configured in config.json) inside the data folder you selected.
   • If photo analysis was run, _photo_inventory_description.txt will also be saved/updated in the data folder.