AI as an API
summary
Combines two UCI spam datasets into a cleaned corpus of roughly 200,000 samples, tokenizes and labels messages, trains an LSTM via Keras/TensorFlow with batch size 32 and 5 epochs, and measures training time across Google Colab, AWS SageMaker (GPU fastest) and an Apple M1 laptop (orders of magnitude slower for LSTM because TensorFlow for Apple silicon is unoptimized). Exposes inference as a REST endpoint using FastAPI and logs predictions to DataStax Astra DB for later human labeling and continual retraining; project is an experiment, not a production system.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:50Spam scale and impact
- 3:16ML spam experiments
- 8:13Training AI model
- 13:07AI model as API
- 14:54Cloud provider lessons
- 19:28Q&A
related talks
41mAI in Web: TensorFlow js
20mAI and Machine Learning for developers
13mAI: Big Expectations
14mAI for a Billion Voices: Taking AI where there is Maximum Impact
38mAI in Kubernetes: How to Get Started?
43mGetting started with AI for free
58mAI-Enhanced Big Data: Integrating Private LLMs and Vector Databases
17mAI in Production: Key Lessons