chandan_maruthi
chandan_maruthi on Hacker News
Not saying it is an achievement. But it a concept I think must be well shared. And if you dont use AI in writing in 2025 you may as well live under a rock.
Let me know what you think
Posted by
chandan_maruthi on Sunday July 20, 2025 @09:11AM
from the self-post dept.
*Distributable Expertise: Why We’re Entering the Age of Shareable “Micro‑Robots”*I’ve been thinking about a new pattern in our community: experts aren’t just writing blog posts or tutorials anymore—they’re packaging their entire workflow into runnable automation templates. I’m calling it *Distributable Expertise*: bundling domain know‑how and code into “micro‑robots” that anyone can import and...
Posted by
chandan_maruthi on Wednesday July 09, 2025 @02:32PM
from the self-post dept.
Most AI agent ideas are lame; they’re essentially a glorified Zapier (I don’t hate Zapier; it has its place). I have been struggling to understand what a good AI agent should look like.I have been using Cursor for a few weeks, and here is why it’s a good AI Agent.* It is a complete experience: Cursor did not integrate into VS Code as a plugin; they just forked the damn thing—owning the coding...
What do you mean by "Now begin littering hyperlinks throughout them"
Posted by
chandan_maruthi on Friday June 06, 2025 @03:36PM
from the ask-hn dept.
We did something interesting recently, but not sure how to proceed. Ideas are welcome.• Context: We (an AI for Support startup) recently onboarded a support team that had no formal KB/HelpCenter. Their “knowledge” was scattered across tickets and Slack threads.• What we did:
– Ran an AI pipeline to convert ~6 months of support tickets into 1700 FaQ/KB-style entries.
— The typical entry was...
Posted by
chandan_maruthi on Wednesday May 28, 2025 @12:22AM
from the self-post dept.
Just realized. Our society values unique takes on subjects. They do not penalize bad takes; both good and bad takes receive the same value.To me, this seems to make sense. If everything is an experiment, then every unique take adds to the corpus of knowledge. Whether a take is good or bad does not affect the value added to the total knowledge.What do you think?
Posted by
chandan_maruthi on Monday April 21, 2025 @01:33PM
from the self-post dept.
Last week I spent three days wrestling with a data‑pipeline script. The API returned pages of JSON that—despite using cursors—felt impossible to checkpoint reliably. Every new edge case sent my code spiraling into complexity.Then I did something old‑school: I grabbed a yellow notepad, sketched the entire flow, and within minutes I had an elegant design I was actually proud of. It was:Simple to...
One more related question, are there any metrics around conversation rates to be aware of.
Posted by
chandan_maruthi on Thursday February 06, 2025 @01:03PM
from the ask-hn dept.
Hi we are getting ready to launch our first self onboarding flow at Twig.so. I love the idea of customers understanding what we do and getting started self serve without a "Book a demo". My big question today is should I
1.) Place a stripe CC paywall with a 30 day trial?
2.) Allow users to try out the product with you using a credit card and then show warnings closer to 30days.I have not done...
"custom dependency resolver" I am intrigued, Ill check it out.
How do you handle dependencies like libraries?
I can definitely relate. As an engineering manager, migration projects are often the least fun to work on. They are complex, take a ton of time and no one wants to work on them. If this works as promised. I would like to try it on a few migrations myself. Best of luck on your work,
Damm!!! I actually thought they shut down.
A an engineer and founder I can say this makes a ton of sense. There are often tons of generic screen very app needs to build that are just not worth the time building custom, I can imagine myself using Second for such screens that are not core to my functionality but essential to complete the product story. Kudos for the work Eric.
Update: Still waiting on OpenAI to increase API limits after we hit hard limits after a popular HN post
Thats an interesting use case. How would you think you would use it on RFCs
Yes the only have a completions api now
Dang, Hi, If you see just below the "add me to waitlist", There are live Chat Bots created for several companies you can try out right now.
These chat interfaces are live and work on custom/fine-tuned models that have been trained on the documentation of these companies.
Over the last 9hrs users exchanged over 6000 messages with the bots. So it's live, it works, and users can try and play with it right now and give feedback.
A few hrs ago the bots hit OpenAis usage limits, Else its was working actively with over several hundred unique users interacting with the bot asking 4-5 questions each
Just saw someone tried to hack/denial-of-serive the bot by just putting too many words in the prompt. (This hit triggers as usage is token based) .
I just set a prompt limit of 100 characters should be plenty. Bot is till down until OpenAI resets limits. https://twitter.com/chandanmaruthi/status/161213449603556966...
Hi All, We had over 6,000k interactions and have hit OpenAIs hard api usage limits :-). I have requested OpenAi to increase limits. Until then the live chat with bot may be restricted. Remember to get on the waitlist with "get access" if you want to give it a try when the live chat starts again.
Stuff that was very hard just a few years ago is now relatively easier to do. I guess that is the nature of technology.
That must have been me :-), I realized the thumbnail and fav icons were not updated so made a push. Sorry about it. I saw people posting on Twitter and I had to update them for meta tags unfurling stuff.
Ha ha ha ! You have a point, That would be the next problem to tackle. Probably try to determine if I should even try to answer a question before attempting it. And then not showing results that don't fit criteria. At a high level, it seems like some embedding works need investigation but I haven't done much evaluation there. There would be filtering at both question and answer levels I suppose.
Good idea, I have not considered on prem yet. But I can see the need in cases where data is sensitive.
Thank you, Yeah thats a good idea, Like literally take everything from the shared drives and train on it.
I am still figuring that out. More updates soon.
Hi, I am training a custom GPT model on documentation published by these companies on their documentation websites. Its not reading from code/comments. But their documentation/support sites.
Hi, If you see just below the "add me to waitlist", There are live Chat Bots created for several companies you can try out right now.
These chat interfaces are live and work on custom/fine-tuned models that have been trained on documentation of these companies.
Over the last 3hrs users exchanged over 1000 messages with the bots. So it's live, it works, and users can try and play with it right now and give feedback.
It should check all the boxes.
The waitlist is to allow us to onboard companies in an orderly way as each bot is trained on their custom documentation.