E-GEO

A framework for E-Commerce Generative Engine Optimization.

---

Setup

1. Clone the repository

git clone https://github.com/psbagga17/E-GEO.git
cd E-GEO

2. Pull the datasets

git lfs pull

3. Install the dependencies

uv sync
source .venv/bin/activate

4. Add API keys

Create a .env file in the project root and include your OpenAI or OpenRouter key:

OPENAI_API_KEY=your_key
# or
OPENROUTER_API_KEY=your_key

Project Structure

E-GEO/
│
├── data/                    # Dataset files, product rankings, experiment inputs
│
└── src/
    ├── all_init_prompts.py      # List of 15 initial prompts used for meta-optimization
    ├── prompts.py               # All system/user prompts used across the project
    ├── prompt_optimization.py   # Main meta-optimization loop (train/validation/test)
    ├── run_experiments.py       # Batch processing, parallel execution, result logging
    ├── analysis.py              # Metrics, histograms, statistical summaries
    ├── utils.py                 # Helper functions
    └── visualiser.py            # Graph plotting utilities

Command-Line Flags

Below are all available flags for prompt_optimization.py, with a clear explanation of what each one controls.

--method Selects which initial system prompt style to use for optimization (e.g., authoritative, friendly, concise, etc.).

--data-path Path to the dataset directory containing all product descriptions, rankings, or experiment data.

--batch-size How many data samples to process per batch during each experiment cycle.

--batch-num Total number of batches to run per epoch for training.
Final number of processed samples = batch_size * batch_num.

--num-epochs How many optimization epochs to run.
Each epoch evaluates all batches and updates the meta-prompt.

--meta-model Which LLM to use for evaluating and optimizing prompts (e.g., gpt-4o, gpt-4o-mini, etc.).

--start-epoch Resume training from a specific epoch.
Useful if you're restarting a long experiment.

--start-batch Resume training inside the start_epoch at a specific batch number.

---

Example Usage

python3 prompt_optimization.py \
    --method authoritative \
    --data-path ../data/ \
    --batch-size 100 \
    --batch-num 10 \
    --num-epochs 5 \
    --meta-model gpt-4o-mini \
    --start-epoch 0 \
    --start-batch 0