rafaepta

not really (just saw Show HN is only for one owns projects). i don't have any credits in this. Edited

measuring humor might be halfway to measuring taste. congrats on this... really original contribution. wonder if you're planning to evolve the benchmark to incorporate a multi-language dimension. would love to see how Mistral and models built outside the US would perform.

I really like the concept, reminds me of dupehound https://github.com/Rafaelpta/dupehound

Litepipe does have PII redaction and privacy controls, but tbh in some regulated industries this might be too early to adopt. Not bcs of the tech but legal and compliance. I don't see regulated industries being the right fit for now.

Thanks, could you elaborate more on how would you use it? Have tested options for capturing context such as Screenpipe and Littlebird?

good one. most likely the capture engine will not even be something that makes sense in 2 yrs. i don't have an answer for this tbh. thoughts?

Thank you, lmk if there is something I can improve

Makes sense. Connecting to an agent is something that already works. The memory is plain SQLite and media files in a local folder on your Mac, so you can point any agent straight at it. I am also shipping an MCP as suggested above in the discussion.

Great ideia, just opened an issue here: https://github.com/Rafaelpta/litepipe/issues/1 will let you know when the MCP is ready

Interesting... How a tool would need to look like for you to use in your industry? Have you looked for note-taking apps before? If yes, why you didn't stick w them?

Yeah, that is pretty much what it does already: it tries to recognize test files and skip them. Dupehound is available for 12 languages Today.

Some languages like RUst you mentioned, have a clear tag that says "this is a test," but others do not, so the tool has to guess from file names and ends up missing some and skipping too much.

Also as I mentioned on the answer below, sometimes you actually do want to see the repeats inside tests, or normal code repeats on purpose too. So I am leaning toward letting users wave off one specific case by hand instead of skipping everything blindly.

Using dupehound for identifying duplicated code.

What I use for: I use for identifying duplicated code. It is deterministic, doesn't use AI, offline, runs from CLI and is super fast (and free).

What I dislike: I won't say it I dislike, but it is not a tool that does all the jobs of a code review. For instance, it doesn't flag security issues. It is superfocused on code duplication (it performs better than Sonar for this use case) and is specifically useful for large codebases. Disclaimer: I am one of the collaborators, so take it with a grain of salt https://github.com/Rafaelpta/dupehound

agreed, but imo terminal and any conversational interface is not a silver bullet. Example: signup up for a ai assistant that lives in the email. Can't cancelled it bcs the founder assumed that since you ask anything in natural language every problem the user might have is solved. This is just pushing problems to users.

So true, just built a deterministic system to identify duplicated code. It's offline and doesn't use AI on purpose, since a gate that blocks your CI has to give the exact same answer every time, and finding dupes means comparing every function against every other (that's index work). It does NOT use AI. But ironically, I used AI to build it (https://github.com/Rafaelpta/dupehound )

I think "deterministic" is going to become a feature label the way "offline" and "no tracking" did.

Great read. Thanks for sharing

watches age well

This article misses the point. It is not about AI replacing workers, but about AI bringing more ROI. Can an AI convert twice as many customers as a $4k salesperson? It is reasonable to say that in a B2C setting, YES. I've seen that. Better SLA, fast responses during weekends, better adherence to existing playbooks, mapping out objections that are not in the playbook, and suggesting updates for the same prompt. In one week, the playbook evolved, and today we are converting more customers than the sales team. Does it capture the value of the $4k usd sales person ? If the ROI is superior, yes. Will I pay for it? That is a different story (we developed this ourselves).

Not really, that is not what we do.

Thanks for flagging this. I do need a better way to explain what we do.

Systems always win. If you don't have anything in place, quick wins are just a distraction.

You’re right. I’ve just changed the title of the article adding “(un)” to the beginning. I can’t prove the errors are intentional, but given the author’s reputation it would surprise me if they weren’t.

I sometimes feel guilty. I’ve tried to set goals (I really have) but it’s just not how I’m wired. I tend to improvise my way through things. Even as a kid, I remember never feeling that urge to "win" at anything. Sports, board games, whatever. Other kids would light up with competition. I’d just… show up, participate, drift through it. I always felt slightly out of sync with that whole dynamic. That’s why this line hit me so hard: “Some of the most powerful forms of progress emerge from people who stopped trying to win and started building new game boards entirely.” Maybe that’s been the point all along. Thanks for sharing this.

I’m betting we’ll all turn into part-time philosophers, staring at the old questions: Who are we? Why are we here? Where do we go next?

Finding meaning becomes the core human task.

Our best tool might be the oldest one: basic ethics. Plato, Aristotle, the $5 paperbacks gathering dust.

I miss Rushmore’s plain approach. Just enough quirk, sharp acting, and visuals that back the plot instead of hogging it. Newer Anderson films look like photo shoots: pretty, but the story drags. Same story dev teams hit when designers chase pixel-perfect screens and users still wait on real features.

ChatGPT ≠ Google-scale.Google: ~14 B searches/day; ChatGPT: ~37 M (~1 : 400). Only ~15 % of ChatGPT prompts look like classic “search”; most are writing/code tasks. Google’s own search volume grew 22 % in 2024 and still holds >90 % share. An LLM citation is nice for credibility, but it won’t move traffic or revenue anytime soon.

Never wondered why 14. But makes perfect sense. Early Unix feels like a masterclass in minimalism under constraint. Also fun detail: Unix V3 had 8-character limits before DOS made it famous.

A few numbers to keep the AEO hype in check:

- Scale gap: Google still processes ~14 billion searches / day; ChatGPT hovers around ~37 million. You’re looking at 1 LLM query for every 400 Google queries.

- Intent gap: Only ≈ 15 % of ChatGPT prompts are classic “informational search.” ≈ 75 % are generative tasks (code, drafts, summaries) that were never clicks in the first place.

- Traffic gap: When Google shows an AI-Overview, downstream organic CTR drops ≈ 70 %. Dedicated AI engines send ≈ 96 % less referral traffic than classic Google results.

Implication: Chasing LLM citations is fine as a branding bonus, but it’s not a growth lever for most sites. Spend effort proportionally.

A pragmatic playbook (what actually moved the needle in my own tests)

1) Question-first headings – treat every H2/H3 like a unit-test: Q in, 40-60 word A out, then elaborate.

2) Answer-first, expand-later – inverted-pyramid writing is a killer for snippet extraction.

3) Chunk + schema – short paragraphs, lists, FAQ/HowTo markup. The clearer the structure, the higher the chance an LLM copies you verbatim.

4) Keep it fresh & sourced – stats with a 2025 timestamp and a citation get surfaced more often (recency + trust signals).

5) /llms.txt – a zero-cost “robots.txt for LLMs” to spotlight the pages you actually want crawled.

Building a QA platform so non-tech folks can test and improve LLM products WITHOUT needing developers.

It’s crazy how much time developers waste tweaking hardcoded prompts, writing endless test cases or constantly shift focus between building LLM features and verifying them. It’s usually the product person who should be handling this.

To scale a product, you need an extra set of eyes to double-check the engineers' work before it goes live.

Just want to give non-technical folks the power to test and improve LLM apps themselves!

This project is called https://ottic.ai/