TODO List for ESP Report Project

Step 1: Project Setup

• Set up virtual environment
• Initialize Git repository
• Add .gitignore file

Step 2: Model Fine-Tuning

• Prepare dataset in JSONL format
  • make a function to take excel into JSONL
    • function that gets the sanitized name equivalent
    • check how many matches we get of the QC data and the extracted text examples that we have
    • find and download missing files (currently 43)
    • scrap text from the finetuning examples
      • formalize the tool that takes the text
• Fine-tune Llama 3.2 model
• Save fine-tuned model in models/ directory

Step 3: Inference Script

• Write inference script
• Test model predictions with sample data

Step 4: Deployment

• Push model to Hugging Face Hub
• Deploy model using Hugging Face Inference Endpoint

Additional Ideas:

• General AI where one person can just ask questions about any of the documents
• General AI where you can ask questions about one document
• How to make general AI from structured data, that can ask questions about any of the wells
  • This is likely something that has already been done
  • How to prevent it from answering questions from other people's wells