Implementing Exercise Wearables Analysis w/Spark and Akka
summary
Built a full exercise pipeline: Pebble accelerometer sends 1.24s BLE packets to an iOS app that fuses watch, phone and heart‑rate streams; core signal processing and Kalman filtering live in C/C++ and Swift, classification started as libSVM but moved partly to device. Server side runs Akka cluster sharding with persistent actors, events journaled to Cassandra and analyzed in Spark + Cassandra for model training. A 100‑user trial failed due to packet loss and bandwidth; solution is a distributed lambda design with on‑device speed layer, feedback and serialized models pushed back to devices.
AI-written summary · about
chapters (14)
- 0:00Intro
- 2:07Pebble Watch
- 4:15Pebble Watch Application
- 9:46Mobile Application
- 12:50Swift
- 18:54Device sessions
- 22:46Upper cluster
- 25:51Profile Service Exercise Service
- 27:59REST API
- 34:09Classification
- 39:57Apache Spark
- 46:32Demo
- 48:27Problems
- 53:22Mobile Applications
related talks
42mSo, You Want to "Do Big Data?"
51mComposing high performance process workflows with Akka Streams
56mAkka in Heterogenous Environments
47mA Frontend Server, Front to Back
47mBuilding High Performance Real-time Stream Processing Microservices
55mExperience report on using Akka in production: pitfalls and rewards
29mLessons from Using Bleeding Edge Tech in a Startup
55mBootstrapping Microservices using Akka, Kafka ans Spark