– An average day has about 20 million API calls, busy days have 50m+

– Breaking down where C, Node and other languages are being used

– What happens when you upload an image to their API

– Really figuring out the file type of something that’s been uploaded

– Dealing with edge cases as they come up but preparing a bit ahead of time

– Switching from Core ML on Apple hardware to Tensorflow on AMD hardware

– There’s no framework sitting on top of the Node API server

– The customer facing web dashboard is using Express, the marketing site is Jekyll

– They’re mostly B2B so feature requests end up being 1 on 1 calls

– Handling payments with Stripe and using a Node / Angular app for it

– Using Lua with nginx for rate limiting, also nginx is their load balancer

– You can’t go wrong with boring and predictable technology

– MongoDB, Redis and Elasticsearch are all running on 3 nodes each

– Having nearly instant access to a ton of data helps figure things out

– What it was like finding a freelance C developer

– Sending webhooks out is controlled by a separate Node Bull driven app

– Dealing with GDPR compliance and storing images on GlusterFS for 1 hour

– Going with bare metal servers in their own data center over the cloud

– The servers have 32-256GB of memory and a range of different CPUs

– Having spare parts and dealing with hardware failures

– About 50 servers run the latest Ubuntu LTS and are managed with Puppet

– The deployment process for a number of different services

– It takes ~30min to replace a drive and every service is tripled up

– The database servers are replicated and there’s alarms and alerts set up

– Rate limiting was put in place for limiting API calls to customers

– There’s custom payment rates depending on each customer’s requirements

– Best tips? Over provision like crazy and monitoring lets you sleep at night

– Do what works for you, don’t copy another company because it works for them

– Check out https://optidash.ai, their tech blog and GitHub account