- The goal of this project is to create a personalized product recommendation system based on user interactions on posts on a social media paltform.
- Users can perform actions such as liking images or uploading images, which will be sent to an Visual Transformer image captioning model to generate captions.
- The nouns extracted from these captions will be stored as user-specific interests. These interests will then be used to generate potential product recommendations using OpenAI.
- The product recommendation list will be displayed to users, and the interests database will be used to show relevant ads to users with similar interests.
- Utilizing the ViT-GPT2 visual transformer model as the core image captioning model for extracting meaningful insights from image captions.
- Implementing user interactions such as image liking and uploading through AWS Lambda for API services, and storing relevant data in AWS DynamoDB for efficient database management.
- Employing Snowpark in Snowflake to store generated captions as User-Defined Functions (UDFs) for streamlined access and retrieval.
- TextBlob for noun extraction from the captions.
- Leveraging OpenAI's LLM API to generate potential product recommendations based on user interests, ensuring personalized suggestions.
- Conducting Exploratory Data Analysis (EDA) on user clicks using plotly, analyzing patterns such as user clicks on specific product categories at different times of the day.
Documentation - https://codelabs-preview.appspot.com/?file_id=18qScGcxVTydmxhaIO8eRmhzS3TcmaczcPK5606xLQdM#0
User Action -> Liked post of the woman sitting on the beach
Image processing -> Fetched relavant Ads
