Ketan Bhatt
@_ketanbhatt
Followers
314
Following
6K
Media
59
Statuses
1K
In hindsight, 2019 was an exceptionally eventful year for me. Traveled to 3 countries, & 3 states in India; took 4 months off; shifted to Dublin, living on my own for the first time; for @intercom, my first ever largeish company. Extremely grateful for this. Thank you, fate(?)!
3
0
19
I wrote about our experience using Elasticsearch for running Fin.
A new entry on our Intercom AI blog from @_ketanbhatt, about our decision to stick with @elastic as opposed to adopting a new VectorDB for Fin. tl;dr: operational knowledge is valuable, and ES will do vectors. Link below!.
1
4
13
RT @jmfryan: I’ve been working on ways to integrate our AI agent Fin into pretty much anything. It means folks can use the full power of F….
0
9
0
RT @isro: Hat-trick for ISRO in RLV LEX! 🚀. 🇮🇳ISRO achieved its third and final consecutive success in the Reusable Launch Vehicle (RLV) La….
0
2K
0
RT @mariokostelac: TIL To figure out the dependency tree for a specific package, is quite faster than reading pip-c….
0
1
0
RT @grandazz: I'm exploring a video project idea for @EmberConf. For it to work, I'll need a bunch of @emberjs users to send me a 2 second….
0
17
0
Finetuning Large Language Models
magazine.sebastianraschka.com
An introduction to the core ideas and approaches
0
0
1
The Definitive Guide to Rack for Ruby and Rails Developers
writesoftwarewell.com
This article explains Rack, including both the protocol and the gem, as well as the concept of middleware.
0
0
2
RT @kevinmartinjos: I just launched . It fetches articles from your twitter feed and lets you read them on your RSS….
0
2
0
Going to lose one hour of sleep to DST, girlfriend suggested I catch up on it in October.
1
0
3
So:.1. If you upgraded to and use the `elasticsearch-ruby` client, please confirm if you got bit the same way. 2. Operational Health Reviews (hat tip @gregolsent) are the real hero of this story, and @brian_scanlan 🤠 7/7.
0
0
6
Anyway, @brian_scanlan realised we had an AMI rollout at 4am on Feb 14th, and the new AMI comes with an updated version of curl. We use the `elasticsearch-ruby` client which uses curl, and we confirmed that downgrading curl to the previous version fixed the issue. 6/7
1
0
3
Result: Teams get a better sense of what "normal" looks like for the infra they own, and can tell if something doesn't "feel right". This bump in the response time was caught by an operational health review! 5/7
1
0
1
@gregolsent did a fantastic job in removing all the friction for teams to carry out these reviews. He created a Google Docs template for them to just duplicate and fill, with guidance and an accompanying Datadog dashboard to answer the questions in the review doc. Result? 4/7.
1
0
2
But if a product team runs Infra that only (majorly) they need , they own the infra (with core-tech's help). We ask these teams to run an "Operational Health Review" every 3-4 months, to avoid silent "drifting into failure" (a term I came across in 3/7.
1
0
3
In Intercom, we have a team that owns Intercom's use of the core technologies we use (so Ruby, Rails, Elasticsearch, MySQL, etc.) as well as shared infra. Ryan wrote about the team in: 2/7.
intercom.com
How does a rapidly scaling company manage the core technologies it relies on? This is how we evolved a dedicated core technologies team.
1
0
2
Short story: On Feb 14th, early morning, something changed. P50(duration) for requests to #elasticsearch increased by 4 times! Thankfully (for debugging), this happened to all our ES clusters. 1/7
1
2
12
