Please enable JavaScript.
Coggle requires JavaScript to display documents.
Engineering systems for realtime predictions in DoorDash - Raghav Ramesh…
Engineering systems for realtime predictions in DoorDash - Raghav Ramesh
Intro
ML looks like
ML is
90% systems
Robust told
Templatizing best practices
10% algorithms
Evolution of ML
Simple rules
Learnt coefficients
First ML model
Many ML models, each with own stack
Full fledged ML
Templatized
Ease of use
Simple tools
Integrate into ecosystem
System
Training
Common features
Standardized
Key - value
Check pipelines
Use historical data
Predictions
Request
Service
Multiple
Low latency
Load balancer
Http
Other details
Monitoring
Response
In action
Critical elements
Accuracy
Debuggability
Time to ship
Dependencies
Custom code
Economies on scale