docs(README): expand with recording, training, and deployment instructions

Author: abrichr

README.md

@@ -10,6 +10,7 @@ OpenAdapter simplifies deploying advanced screenshot parsing and action models o
 - **Cost-Efficiency**: Deploy high-performance instances on demand, with intelligent caching and resource pause/stop features to reduce costs.
 - **Container & API Compatibility**: Supports Dockerized models like OmniParser and Set-of-Mark, with future support for Anthropic and OpenAI APIs.
 - **CI/CD with GitHub Actions**: Automated integration and deployment ensure consistent updates.
+- **Dataset Preparation and Fine-Tuning**: Collect and fine-tune your models with OpenAdapter’s tools for recording, preparing, and training models using user interaction data, such as screenshots and actions, captured directly in your application.
 
 ### Prerequisites
 - **Python 3.10+**
@@ -41,9 +42,41 @@ OpenAdapter simplifies deploying advanced screenshot parsing and action models o
    ```
 
 ## 💡 Usage
-Straightforward commands to deploy and manage model instances.
+OpenAdapter provides commands to deploy and manage model instances and capture user interactions for fine-tuning. You can record, train, and tune models with your custom dataset, all managed within the OpenAdapter environment.
 
-### Example Deployment Script (OmniParser)
+### Recording User Interactions
+To capture user interactions such as screenshots and actions, use OpenAdapter’s `record` command:
+```bash
+python -m openadapter.record "doing taxes"
+```
+This command will save the actions in a database file:
+```plaintext
+Actions saved to ~/openadapter/recording.db
+```
+
+### Preparing and Fine-Tuning the Model
+Use the recorded data to prepare a dataset and fine-tune your model:
+
+1. **Prepare Dataset**:
+   ```bash
+   python -m openadapter.train.prepare ~/openadapter/recording.db
+   ```
+   Example output:
+   ```plaintext
+   Preparing dataset from: ~/openadapter/recording.db
+   Dataset prepared at ~/openadapter/prepared_data
+   ```
+
+2. **Fine-Tune Model**:
+   After preparing the dataset, specify the paths to the prepared data for fine-tuning:
+   ```bash
+   python -m openadapter.train.tune --caption_model_path=~/openadapter/prepared_data/caption_model --som_model_path=~/openadapter/prepared_data/som_model
+   ```
+
+This flow enables OpenAdapter to use custom datasets created with OpenAdapt for more accurate action detection and screenshot parsing. Adjust paths based on your local setup.
+
+### Deployment Example (OmniParser)
+Deploy OmniParser using an AWS GPU instance with OpenAdapter:
 ```python
 from openadapter.server import OpenAdapterConfig, Deploy
 import fire
@@ -95,7 +128,37 @@ python -m oa.deploy ssh
 - **Planned**: Hugging Face, Anthropic, OpenAI; future support for GCP and Azure.
 
 ## Integrations
-Works with OpenAdapt or as a standalone solution for automated model deployment.
+OpenAdapter works seamlessly with OpenAdapt to build datasets and automate models. It can also function as a standalone solution for deploying and managing models in automated environments.
+
+## Requirements
+
+### Core Requirements
+- **Python 3.10 or higher**
+
+### Optional Components
+Install specific dependencies based on the use case:
+
+1. **Recording**: Required for capturing user interactions.
+   ```bash
+   pip install openadapter[record]
+   ```
+
+2. **Training**: Includes dependencies for preparing datasets and fine-tuning models.
+   ```bash
+   pip install openadapter[train]
+   ```
+
+3. **Deployment**: Necessary for deploying models to production.
+   ```bash
+   pip install openadapter[deploy]
+   ```
+
+4. **Full Installation**: Installs all dependencies for full-featured use.
+   ```bash
+   pip install openadapter[full]
+   ```
+
+> **Note**: GPU support is recommended for training and fine-tuning tasks, especially when working with large models like YOLO and BLIP2.
 
 ## 🛠️ Roadmap
 - **AWS CDK Automation**: Streamline Infrastructure as Code.