Our Food Recommendation System leverages advanced machine learning algorithms to provide personalized food suggestions to users. The system is built on a robust architecture that processes data, trains recommendation models, and serves these recommendations through a user-friendly interface.
The core of our recommendation engine, where data is processed and machine learning models - Jasper and Marlin, are trained to generate food recommendations.
- Apache Airflow
- Pandas
- Numpy
- Wandb
- Slack Integration
- Amazon S3
raw: The raw input data retrieved from our databases. This data forms the basis for our recommendations.processed: Data that has been transformed and enriched by our scripts to be used for training our models.
Files:
models.pkl: Serialized versions of our trained machine learning models.model_metrics.csv: Performance metrics of our models, used to track and improve their accuracy.
Apache Airflow manages our workflow orchestration, ensuring that our data pipelines are executed in a consistent and reliable manner.
A collection of scripts that automate routine tasks such as:
- Downloading data
- Saving model states and results
- Inserting schedule results into MongoDB
- Shutting down machines after processes are complete
Our data processing and machine learning scripts run in an isolated Python virtual environment (denoted as 'jasper') to maintain dependency integrity.
A front-end application, named after our Marlin model, which interacts with users and displays the personalized food recommendations generated by our system.
The runtime for our server-side logic, which interfaces with MongoDB and serves the Marlin front end.
A NoSQL
database used for storing and retrieving data efficiently. It plays a crucial role in managing user profiles and recommendation data.
A lightweight component of our system responsible for specific tasks that require minimal processing power.
Our secure database where sensitive user data is anonymized to protect privacy while still allowing for effective data analysis.
A relational database that stores structured data, which might include user information, food items, and other metadata essential to the recommendation process.
Jasper is a machine learning model dedicated to understanding user preferences and dietary requirements to suggest new and exciting food options.
Marlin complements Jasper by optimizing the suggestions for nutritional balance and user satisfaction, using a more extensive set of data points.
- Data Ingestion: Raw data is ingested from MongoDB and the Masked DB, ensuring a rich dataset for processing.
- Data Processing: Util Scripts download and process the data, making it ready for use by the machine learning models.
- Model Training: Jasper and Marlin models are trained in the Python virtual environment using the processed data.
- Recommendation Generation: Models generate food recommendations, which are then stored in MongoDB.
- User Interaction: Users interact with the Marlin Front End, where they receive personalized food recommendations powered by our models.
- Feedback Loop: User feedback on recommendations is fed back into the system, allowing Jasper and Marlin to learn and adapt over time.
For a more detailed explanation of the data flow and system operations, refer to the architecture diagram included at the top of this document.
Made with ❤️ by Asad
MIT © Asad