Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
bosky101
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
7 ms
·
61.
▲
by
bosky101
2y ago
wrt (3) Being able to self-host is extremely important. I noticed a lot of focus on the docs on the Quickstart/client usage. But things like default storage engine as a ENV, path for storage as an ENV. These are very important. hmm. Re
62.
▲
by
bosky101
2y ago
We have never met. But great to see you here!
63.
▲
by
bosky101
2y ago
The website and example, and usage looks clean. Kudos! I have some questions around what's happening under the hood that werent evident from an initial read of both your website as well as GitHub. 1. Does subscribe listen for new chang
64.
▲
by
bosky101
2y ago
Some questions remain answered from your home page ( from someone at a mid sized firm writing their "build-it-yourself" version of the same goal at largish scale) 1. How often is each campaign run? 2. How does it avoid re-sending
65.
▲
by
bosky101
2y ago
before and after preview => https://i.ibb.co/X23D18k/zunder-before-after.png
66.
▲
Turbo charge your Mac(zsh) in 5 lines
1 points
by
bosky101
2y ago
|
2 comments
67.
▲
by
bosky101
2y ago
Over a weekend last month - at an AI hackathon made https://nonfungibledocs.com Create variations of your content, so that you can backtrack the source of a leak. Just changing 5 words in various places of a page of a doc, it ca
68.
▲
by
bosky101
2y ago
I haven't even got to kinesis or bandwidth or storage. Even if you compress N objects through spark/etc your starting point would be the large number of writes first. So that doesn't change. The costs would be even larger con
69.
▲
by
bosky101
2y ago
I didn't realise this was meant for consumers. To make this a B2B play - consider allowing youtubers to paste a url or video to create a pop quiz & course. Upto N slides / chapters freemium.
70.
▲
by
bosky101
2y ago
Not sure of the CH Kafka engine but generally I think you should partition by userId. Because the next step would be trying to run some cron for a user or event based trigger based on the events. And the only way to avoid multiple machines
71.
▲
by
bosky101
2y ago
Storing small events in s3 can explode costs quickly. At 1M events/day that's $7.5/day. Decent At 15M, $75/day Cost for 150 million S3 PUT requests per day of 25KB each would be $750/day, assuming no extra data tran
72.
▲
by
bosky101
2y ago
1) Appreciate the single image to get started, but am particularly curious how you handle different events of a new user going to different nodes. 2) any admin interface or just the rest API? 3) a little bit on the clickhouse table and engi
73.
▲
Over 800 GenAI Hackathon Ideas
(devfolio.co)
2 points
by
bosky101
2y ago
|
0 comments
74.
▲
Show HN: Roast/review my hackathon submission in a microniche like doc security
(devfolio.co)
1 points
by
bosky101
2y ago
|
0 comments
75.
▲
Power of Micro Data Centers
(von-neumann.ai)
1 points
by
bosky101
2y ago
|
0 comments
76.
▲
by
bosky101
2y ago
stumbled on the builder's blog, then wobaka earlier this week. this guy is on a roll! btw, the 'book demo' CTA on wobaka, should go to one of these forms.
77.
▲
by
bosky101
2y ago
That time of the year when everyone can see there is craftsmanship in the tiny details. https://littlebigdetails.com is exactly that
78.
▲
Group buying for hosting, for cheaper large company pricing
(pump.co)
1 points
by
bosky101
2y ago
|
0 comments
79.
▲
President Mode
(twitter.com)
1 points
by
bosky101
2y ago
|
0 comments
80.
▲
by
bosky101
2y ago
Dayum
81.
▲
by
bosky101
2y ago
Try the following - port something you know well into a different language - try to build something open source with 0 deps - a sub 10k website - I heard Chrome manifest version is shuttering, and there are lot of opportun
82.
▲
by
bosky101
2y ago
Spot on. But I think there is also going to be a massive impact on hiring and teams. This ladies and gents, is where we won't need to hire people. Software 2.0 could use the same infra, rely on 100 different datasets and produce 100 ne
83.
▲
by
bosky101
2y ago
Wow. Also happy that highscalability is still up and about.
84.
▲
by
bosky101
2y ago
https://tv.algora.io/docs/streaming/quickstart
85.
▲
by
bosky101
2y ago
Maybe a potential acquirer could start a video team to catchup with modern times.
86.
▲
Hermes: A Text-to-SQL Solution at Swiggy
(bytes.swiggy.com)
3 points
by
bosky101
2y ago
|
0 comments
87.
▲
by
bosky101
2y ago
I spent 10 mins on your docs, website. Excellent effort overall. But I didn't know until I read your comment here about the uploading CSV, instant parsing that non technical people may find very interesting. This is something pgweb for
88.
▲
by
bosky101
2y ago
This should do one thing. Use sizeup or window managers instead
89.
▲
by
bosky101
2y ago
One way to implement this would be to have a checkbox to ignore passwords/single words When checked ignore anything that has a space Slick UI. Well done.
90.
▲
by
bosky101
2y ago
Wow! Well done
More ›