For my UBC Engineering final year project, our team built a Contextual Inference Engine. This SaaS platform analyzed unstructured user content (text and images) to extract metadata—Time, Location, and Sentiment—to serve highly relevant advertising parameters.
Technical Architecture#
The system was designed as a modern microservices application.
1. Model Ensemble#
We didn’t rely on a single model. Instead, we orchestrated a pipeline of various AI services and custom models:
- Third-Party APIs: Integrated Microsoft Azure Cognitive Services, Amazon Rekognition, and IBM Watson for robust baseline analysis.
- Custom Models: Built specialized models using Google TensorFlow and SyntaxNet (for dependency parsing) to refine sentiment analysis and entity extraction.
2. Cloud Native Deployment#
- Containerization: All services were Dockerized for consistency across development and production.
- Orchestration: Deployed on a Kubernetes cluster to manage scaling and service discovery.
- CI/CD: Configuring TravisCI for automated testing and deployment pipelines.
Outcome#
We successfully built a working prototype that could take a raw image or text snippet and output structured JSON targeting data (e.g., “User is happy, at a beach, during sunset -> Suggest Sunscreen or Travel Ads”).
Tech Stack:
- ML Frameworks: TensorFlow, SyntaxNet
- Cloud Services: Azure, AWS, IBM Cloud
- Infrastructure: Docker, Kubernetes, TravisCI
- Languages: Python
Note: The demo is offline as the Azure student credits have expired.
