Show HN: Morning Brief – Track any topic on HN, Reddit and others

38 points by jv22222 ↗ HN
TL:DR - https://morningbrief.ai/setup

Hi, HN --

We're two indie cofounders who bootstrapped a new product to scratch our own itch: there's so much great content out there, but it's hard to find it. Sites like HN serve us like-minded people, but there are many more topics in other non-HN niches.

Our idea is this: an email you get once a day, filled with only top articles (posted to HN, Reddit, etc) that match your specific interests. You can specify which interests, and how many links you want. Ranked by our signal score.

<tech-info>

To start, we built an ingestion engine. In the first few weeks we only processed HN posts (1k/day) then we added Reddit posts (1m/day) then we added some other sites like Indie Hackers and relevant Twitter accounts.

Then we added semantic tagging. To start we used a generalized tagging service, but then quickly realized we needed to build our own layer because they don't understand many of the current trendy topics.

Next up we wanted to add thumbnails to worthy articles (about 10k/day). It became apparent this was going to cost over $1k/month using a commercial API. So we built our own service using headless chrome, Laravel & puppeteer.

Our original plan was to just show headlines, HN-style. But after initial feedback users asked for article summaries. To start we used a 3rd party commercial summarizer. Aside from the cost (appx. $400/month for our needs) we were not super happy with the 3rd party results so we built our a summarizer.

Here's an example of a brief with summaries - https://morningbrief.ai/sent/24960b76-4d494be1-ae822280-14274649 - this example is crypto focused with no thumbnails (keep in mind every brief is 100% custom). Here is another example of a more eclectic brief without summaries - https://morningbrief.ai/sent/377f08da-18f04407-a655d594-b1b3043b

Our stack right now is Python for link processing and summaries, Laravel to grab screenshots, send emails, and be the web app. We use SaltStack to orchestrate infrastructure. We run 13 servers on AWS with a cost of approx. $300/month.

</tech-info>

When we started out we thought this would be a weekend project! Turns out solving timely, personalized, summarized, quality controlled, content aggregation at scale (in a cost effective way) is quite a difficult problem. That's how we managed to soak up 6+ months of our side project time just getting it ready for this Show HN!

Our early users have told us that we've sent them important articles that they missed -- which is exactly why we built this.

You can check out how the brief builder works (without registering): https://morningbrief.ai/setup. We're still experimenting with pricing, but we're trying a 2-week free trial, then $5/mo (annual).

Any feedback would be appreciated!

Justin & Joe

21 comments

[ 3.4 ms ] story [ 60.3 ms ] thread
(comment deleted)
Co-founder here happy to answer any questions!
You guys have built quite a nice and useful product. I was thinking of a similar project which required crawling all of reddit. But presently there are some hiccups due to the large amount of data involved.

I wanted to ask how you guys manage the reddit data? Do you use praw/pushshift to get all the comments/post titles or is it limited to selected subreddits? Including all comments and all subreddits is quite a lot of data I guess. Any suggestions to collect that data and then work with it effectively? Thanks.

We use praw, it works great if you fill out Reddit's forms to register your user-agent as per their terms of service. We get data about all posts, but don't get the comments themselves, just the post metadata (titles, likes, number of comments, etc). Good luck!
Is there somewhere I can suggest sources? I really love your product, but I couldn't see where you got your data from.

Also a small but annoying thing, when I delete topics the confirmation box doesn't respond to the enter key, only the mouse.

Anyway, those are relatively minor issues and I definitely look forward to my daily brief tomorrow.

Thanks so much for your feedback. Oh yeah I hate it when you can't do stuff by keyboard. I'll have a look at fixing.

Re sources. Right now it is select Twitter accounts, all of Reddit, Hacker News, Indie Hackers.

Some we are interested in is github and dev.to - just need to wrap our heads around the details.

What ideas do you have for sources?

This is a good effort. The tech stack seem to work well and results are mostly on topic.

Some results are off though, for example this article https://morningbrief.ai/l2/0/h64qAA showed as recommendation for 'search engine results page'. Perhaps instead of a tagger consider using NLU models (embeddings).

Some recommended topics have not results like 'life extensions'.

It sometimes shows results 6+ days old - not sure why if the goal is to have a daily brief (so should be around 1-2 days old at most).

I like that you have a paid plan from start, but would probably hold off subscribing until I can get at least 3 month multi-topic trial to see if it is worth it first.

Currently the weakest side of the service is design and UX.

3 months free trial! Holy shit we’ll go bankrupt. That said, we’ll probably go bankrupt anyways ;)
‘Destiny convergence’ heh. Can you not bring the cost down to $50/mo until you reach product market fit?
(comment deleted)
Oh BTW I forgot to say. For topics like "Life Extension" it's true that there are not a lot of links. But that doesn't matter too much.

Just add it to your brief and when something pops up you'll get it.

I have about 5 very low traffic topics in my brief, and it's actually quite nice when they show up from time-to-time. Kind of like a bonus.

I wanted to give this a try, but as soon as I type any keyword or select an entry from either of the drop downs, I get a huge "500 Internal Server Error" dialogue box.

Using Chrome on Android

Really weird, I can't seem to find a way to break it. How about Firefox?
Is there a way to export the feed to a RSS reader? Maybe 1 post per section?
Is that something you would pay for? :)
I wish there was an option to log in with my account and scan my preference based on my likes.
Can you clarify that a little bit I don't quite get it! :)
Very sleek. Do you use click through rate as a metric for ranking? If yes, then I fear it is going to promote click bait kind of title/summary/cover picture. If not, why are you redirecting for every links?
For the moment redirecting for every link to get a sense of if the customers are using it. Very early days.
I use Huginn + Google Alerts for this, but haven't found anything great for curating conversation from the Twitter community. For example, show me what was talked about the most yesterday by Space Twitter.
Nice. Do you do any kind of quality control ranking with your setup?