Hacker News: mrr7337

API errors may be worse for retention than crashes

mrr7337 — Sat, 27 Jun 2026 06:40:32 +0000

Article URL: https://rejourney.co/engineering/2026-05-18/ambiguity-kills-app-growth

Comments URL: https://news.ycombinator.com/item?id=48695763

Points: 2

# Comments: 0

New comment by mrr7337 in "Australia plans to strengthen under-16s social media ban"

mrr7337 — Sat, 27 Jun 2026 06:26:21 +0000

Nothing Proton VPN can't solve

New comment by mrr7337 in "Un-0: Generating Images with Coupled Oscillators"

mrr7337 — Thu, 25 Jun 2026 20:54:49 +0000

I didn't really understand anything...lgtm

Recording 10k user sessions cost us less bandwidth than a single HD video

mrr7337 — Thu, 25 Jun 2026 20:48:26 +0000

Article URL: https://rejourney.co/engineering/2026-05-06/mobile-session-replay-cost

Comments URL: https://news.ycombinator.com/item?id=48678988

Points: 6

# Comments: 0

New comment by mrr7337 in "Show HN: A GitHub app that suggests code fixes for conversion failures"

mrr7337 — Sun, 21 Jun 2026 15:31:24 +0000

Great Question! We have a few options when it comes to how we process a 100s of session replays that each can be a few MBs or greater. The main concept here is that we initially use a grouping system that groups the same signals (such as rage taps or dead taps) that occurred on the same app or website page for many sessions, and admit a random sample of that group into deeper analysis, which is an LLM of choice. We also have other tricks up our sleeves such as reusing session context we already put the effort into processing before among other things. We're working on improving this system though so let us know what suggestions you have!

Show HN: A GitHub app that suggests code fixes for conversion failures

mrr7337 — Sun, 21 Jun 2026 12:42:52 +0000

Hi Hacker News!

I’m Rashid, a first-year student at UT Austin and the person leading development of Rejourney, a web/app analytics and session replay tool.

Recently, a handful of our users emailed us that they only care about a few user recordings (replays) that show conversion issues, and didn't want to waste time watching all the session replays they got. And no matter how many filters/smart capture rules we put in, the deterministic methods of choosing what to capture always missed important session recordings or overshot and recorded everything. Our users also used other platforms like Clarity or Posthog's AI that summarized all the replay issues, but those also either missed important sessions or focused on a few sessions rather than the overall picture in terms of revenue, growth, and conversion leaks.

We started to think about the unit of analysis and segmentation strategies for failure patterns across many sessions. That led us to build Marlin (big fish), a GitHub app that looks for recurring conversion failures, connects them to session, console, network, and repository context, and suggests a code change for a developer to review. The cool thing is that if you setup the analytics properly, it will understand which sessions have the "leaks" that impact business metrics and growth significantly. The link is a public demo using a simulated leak detection over a set of replays.

It's still pretty early and what we are trying to understand is how well does this perform. Theoretically, it has all the context and feedback it could possibly get in the best possible translated forms for it to understand from the App/website code to the actual user sequence and recordings.

Let us know what y'all think!

Comments URL: https://news.ycombinator.com/item?id=48618445

Points: 8

# Comments: 2

PostHog training on end-users; Indie Alternatives?

mrr7337 — Thu, 28 May 2026 18:28:50 +0000

I woke up today to one of the craziest attempts to sugar coat AI training. Posthog wants to train on developer data which means MILLIONS of end-users' session replays, interaction data, metadata, etc. What's worse is they were trying to sugar coat it with their "funny" emails (even the subject line was hella sugar coated)

Comments URL: https://news.ycombinator.com/item?id=48313338

Points: 3

# Comments: 2