MLOps End-to-End Speedrun: Build an APIs around a Spam Classifier Model
summary
A working pipeline trains an LSTM recurrent neural network on extracted features, stores models with explicit versioning, and serves them behind Python FastAPI endpoints with route factories so v1/v2 coexist; a cache layer reduces latency. Features are managed via Feast + Cassandra (offline + low‑latency online stores) with a push source for real‑time inserts, and an event‑stream option using Apache Pulsar replaces the inbox API; full reproducible repo is provided.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:10Envelope challenges
- 4:42Apache Cassandra and Feast
- 7:33Raw data to model training
- 9:19Microservice production architecture
- 10:01Model versioning in APIs
- 11:26Feature store and real-time flow
- 12:51Event streaming architecture
- 15:00Closing
related talks
16mProduction MLOps for Spam Detection: Real-Time Multi-Modal AI
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives
50mBuilding an MLOps strategy from the ground up
31mMLOps In Practice – How To Run Your Machine Learning Models In Production At Enterprise Scale
37mAutomated Machine Learning: Building a Pipeline
31mApplied MLOps to Maintain Model Freshness on Kubernetes
13mMLOps: From AI Research to Production-Ready Systems
14mPerformance Optimization in High-Volume Transaction Systems