reacharavindh

HTML first, minimal JS, not using any frameworks unless absolutely justified is already part of my prompts to LLMs when I build stuff for myself. I’m not a frontend developer, and have not kept up to date with the ton of frameworks. That HTMl first approach already makes the LLM outputs faster/fresher for me. I wish more people went this route for public facing services.

I’m using Claude with a $100/month subscription. I’m playing around with using Opus as the Architect, Sonnet as the implementer/engineer and Deepseek-pro as the deep reviewer, and tester. It’s been quite good as I expected. If my usage pattern holds up, I would downgrade my subscription to the $20/month one and toss more money to Deepseek.

Repo reference here: https://github.com/aravindhsampath/agentic-template

As unlikely it is to happen at scale, as a thought process - what would happen if people start selling those index funds in a mad rush? Just drives the transaction volume because those with that new money will just buy something else in the market?

I know SpaceX, Anthropic, and OpenAI will probably be a drop in the bucket in terms of scale of these funds, (free float % etc). But, is it realistic to take the money out of index funds for a bit until the price of these new stocks come crashing eventually?

It certainly is true that many SaaS subscriptions can be replaced with custom written(supercharged by AI) tools. The key question however remains “Who are you going to call when something goes wrong?” Or “who is going to maintain this software?” Because essentially that’s what SaaS subscriptions genuinely offloaded and let you focus on your core business things.

As is most things in engineering, this is a trade-off and the line where buy vs build exists somewhere to be found and justified. Sure AI moves the line closer to build these days.

It will be an interesting vector for competition to attract customers.. when you search on {Claude, ChatGPT, whatever}, you get results based on what the model evaluated, and not based on advertisers who paid the most..

I thought Google would use their playbook of search - initially, everything works so good like magic until they enshittify the service by auctioning the results to who pay the most. This time around they are so bold they start from shit..

I tried to use Zulip as _the_ interface for my personal AI assistant because its topic based memory model fits nicely for a personal assistant. But, the iOS app was barely useful to use daily. It lacked even basic features like the ability to create a topic/channel. I could not believe it would be the case and to my surprise, it was.. I have moved on to choose Rocket.chat for my needs. Choosing to implement E2E notifications over basic functions of a mobile app reinforces my decision to move away.

Sorry for the tone. It was a rant not to sound like an unthankful freeloader, but highlight the outsider perspective. The project can prioritize what they want of course.

The part that makes ipadOS feels like a toy is Apple’s iron grip in “App Store only” app delivery. The thing has so much power, but to do anything useful, you gotta play by Apple’s rules. All that dream of quirky, useful, innovative ipadOS leveraging apps would show up if they relinquished that software control and let indie or otherwise apps get there without the Apple Tax both in money and rules.

Would the iPad still be that days long, cohesive device is another story.. it Apple cannot have their cake and eat it too.

This specific form may be a joke, but token conscious work is becoming more and more relevant.. Look at https://github.com/AgusRdz/chop

And

https://github.com/toon-format/toon

I’m building HobbyBoard - A private(self-hosted) visual library(A Pinterest alternative).

Mostly for myself to use for my hobby. Sharing with everyone because I find it genuinely useful.

Yes, it is coded with assistance of LLMs, but I care for the details and it is not vibe-coded in hours.

https://hobbyboard.aravindh.net/

GitHub: https://github.com/aravindhsampath/hobbyboard

Demo (resets every hour): https://demo.hobbyboard.aravindh.net/

Almost ready to do a show HN :-)

+1 to amplify the voice that hates a subscription to a taskbar. If it was €15 one time I would’ve instantly bought it.

Reminded me of this from my bookmarks.

https://github.com/chr15m/runprompt

I mean it is a business after all, trying to make money..

I must say, the new smart bricks with all sorts of sensors(color, gyro, distance etc) triggers the inner child in me. I can’t wait to get them for my kiddo and teach him how that magic actually works beneath.

The regular LEGO at this points feels “just plastic” and I won’t feel bad offloading that purchase to AliExpress.

Or may be an Open Slack that is as good as Slack :-)

On these lines… it has been on my “to be vibe-coded” list to make an extremely minimal node-exporter(the metrics collector for Prometheus) in rust (support only Linux, gather extremely minimal set of metrics) that uses tightly controlled concurrency so as to fetch all metrics within a short span of time.

If anyone has more AI tokens or spare time with mental energy to burn… go for it :-)

If this makes people develop stuff under the assumption that the user only has 8 GB of memory, I am happy for where we are going :-)

I mean, wouldn’t an even slightest clever adversary do just that? Launch a bunch of decoys(party balloons, drones, whatever) amidst one they actually want to actually use as an attack vector? Would the pentagon just play an expensive pew pew and potentially miss the real one?

I dont. I use this as my coding harness (replacement of gemini-cli/claudecode etc). I dont want to sandbox it because I expect it to be used only for coding on projects. I dont want to over complicate it.

I am building my own assistant as an AI harness - that is definitely getting sandboxed to run only as a VM on my Mac.

A few things. I intentionally clone the repo and build it locally for my use and use it as my-omp.. this way, I can make oh-my-pi make customisations like skills, tools, anything and yet retain the ability to do a git pull from upstream with cherry picking if necessary.

I have this in my shell rc.

   # bun

   export BUN_INSTALL="$HOME/.bun"

   export PATH="$BUN_INSTALL/bin:$PATH"

   alias my-omp= "bun/Users/aravindhsampathkumar/ai_playground/oh-my-pi/packages/coding-agent/src/cli.ts"

and do

1. git pull origin main

2. bun install

3. bun run build:native

every time I pull changes from upstream.

Until yesterday, this process was purely bliss - my own minimal custom system prompt, minimal AGENTS.md, and self curated skills.md. One thing I was wary of switching from pi to oh-my-pi was the use of Rust tools pi-native using NAPI. The last couple of days whatever changes I pulled from upstream is causing the models to get confused about which tool to use and how while editing/patching files. They are getting extremely annoyed - I see 11 iterations of a tool call to edit a damn file and the model then resorted to rewriting the whole file from memory, and we al know how that goes. This may not be a bug in oh-my-pi per se. My guess is that the agent developed its memory based on prior usage of the tools and my updating oh-my-pi brought changes in their usage. It might be okay if I could lose all agent memory and begin again, but I dont want to.

I'm going to be more diligent about pulling upstream changes from now on, and only do when I can afford a full session memory wipe.

Otherwise, the integrations with exa for search, LSP servers on local machine, syntax highlighting, steering prompts, custom tools (using trafilatura to fetch contents of any url as markdown, use calculator instead of making LLM do arithmetic) etc work like a charm. I haven't used the IPython integration nor do I plan to.

I began with pi, and have been using oh-my-pi the last two weeks.

https://github.com/can1357/oh-my-pi

More of a batteries included version of pi.

Thanks for this! I’ve been looking for a better solution to the .env files and this is ideal, covers all my needs.

I am building Hobbyboard as a self hosted visual archive that uses vision models to curate inspiration media for hobbyists and makers.

Website: https://hobbyboard.aravindh.net

GitHub: https://github.com/aravindhsampath/hobbyboard

I want to do a show HN later this week.. but here might be a softer launchpad :-)

To me, the magic is around interactions with personal info where they are - iMessages, e-mails, etc. I still am wart to open up like this, but it certainly is not as simple as Claude code and cron task. The “you can already do this via rsync + FTP” comment on Dropbox Show HN thread comes to mind.

My wife has a Gemini AI pro Ultra whatever that she paid for to help with her academic research, but uses barely. She was generous to login with it on my computer to let me use the cli with her credentials. So Gemini-3.0-pro comes from there. I use OpenAI and Claude API keys with EUR 20 credits that I topped up (pay per use). I've mostly settled on Gemini-3.0, but occasionally pit codex, Gemini and Claude at each other at a gnarly problem reviewing each other's work, and proposing actionable changes with clear reasoning. They tend to catch teachers blind spots, an dI get to learn something from their reviews as well.

I've been thinking about this very thing the last few days. "secretary in my Mac" to be specific. An ever running daemon that uses an LLM model for smarts, but pretty much do as many dumb things deterministically as possible. 1. Fetch my calendars(Fastmail, work Google Calendar, Couple's calendar at Cupla) and embellish it with routine tasks like pickup/drop kids, and give me a Today view like this https://zoneless.tools/difference/london-vs-new-york?cities=...

2. Access to my TODO list on Apple Notes and basically remind my ADHD brain that I ought to be doing something and not let it slip because it is uninteresting.

3. Have access to all models via API keys I configure and maintain a "research journal" of all the things I go to LLMs for - "research of bike that fits my needs" whatever and figure out if there needs to be a TODO about them and add if I say yes.

4. View my activity as a professional coach and nudge me into action "Hey you wanted to do this at work this year, but you haven't begun.. may be it is time you look at it Thursday at 3 PM?"

5. View my activity as a mental health coach and nudge me like "hey you're researching this, that and blah while X, Y and Z are pending. Want me to record the state of this research so you can get back to doing X, Y and Z?" or Just talk to me like a therapist would.

6. Be my spaghetti wall. When a new idea pops into my head, I send this secretary a message, and it ruminates over it like I would and matures that idea in a directory that I can review and obsess over later when there is time..

As you see, this is quite personal in nature, I dont want hosted LLMs to know me this deeply. It has to be a local model even if it is slow.

Using a coding agent over days on a personal project. It has made me think

1. These llms are smart and dumb at the same time. They make a phenomenal contribution in such a short time and also do a really dumb change that no one asked for. They break working code in irrational ways. I’ve been asking them to add so many tests for all the functions I care about. This acts as a first guard rail when they trip over themselves. Excessive tests.

2. Having a compiler like Rust’s helps to catch all sorts of mines that the llms are happy to leave.

3. The LLMs don’t have a proper working memory. Their context is often cluttered. I find that curating that context (what is being done, what was tried, what is the technical goal, specific requests etc) in concise yet “relevant for the time” manner helps to get them to not mess up.

Perhaps important open source projects that choose to accept AI generated PRs can have such excessive test suites, and run the PRs through them first as a idiotic filter before manually reviewing what the change does.

It is always a trade off of fighting spam versus actually communicating. Do I want this person to easily write to me without a lot of effort? Or do I pay the cognitive cost for my audience for the slight chance of win against spammers? I chose to stay simple.

Bingo Bingo Bingo.

I learned of a name to how I have been at 32 years of age, and got formally diagnosed at 39. The trigger was my 6 year old acting like I did when I was 6 and seeing how much of a struggle it is for him and others around him.

Been titrating medication and reworking how I approach my work and personal life the last few months.

The thing about constant stream of ideas, micro ideas, while life pulls in different directions- kids, partner, social life, home needs etc is the struggle I am working on to manage better.

It sounds like we could share notes. Would you be interested to communicate over a private message? I’m based in the Netherlands.

P.s. I’ve got nothing to sell, influence or creep about. Genuinely someone on the same boat and I thought it would be nice to communicate with someone who can relate.

If you’d prefer - my email is aravindh at fastmail com

I had my phase of having a fancy shell with all bells and whistles. I’ve now settled with the default terminal so that 1. It is fast, 2. Whereever I work, I have the same experience.

Throwing stone from a glass box eh? If I understand correctly, US is by far the largest services exporter to EU… should EU merely apply the same “tariffs” that US might impose on these goods, some healthy European alternatives would finally gain some ground..