ramshanker

Personal Context: For our personal ClockStack[0] project, I have decided that I will NOT read code at all. Though at high level, I do decide the architecture, dependency libraries etc. Sometimes even telling database schema. That's it. No code reading. Even ask AI how you implemented is enough. And going for prod/live just by chatting with AI is totally fun. Background: I wanted to do this ~5 years back and read almost 500 pages of H.264/RTMP spec/documentation. Even got valid non-compressing video encoding working in Golang. Then priorities changed. Now with the power of AI: Live within a week. I will absolutely call it vibe coding. It is totally fun.

[0] https://www.youtube.com/@TheClockStack/streams

Just as I was getting comfortable with 10+ days of uptime, we get a neighbourhood internet downtime! It it’s back up. It was a good test of auto resume capabilities I built into my custom streaming application.

Time to build the failover!

Speaking of clocks, recently we have started streaming live clocks on Youtube. There could be 100s of way of showing beautifull clocks on screen. This one is mine. Coding for fun! A bandwidth hog to be honest. ;)

https://www.youtube.com/@TheClockStack/streams

I am in the process of setting up a demo local-hosted-model using either vLLM or SGLang for my not-in-software-business enterprise. What harness would you recommend? I mean what are the top 5 open source ones. Trying to avoid unsustainable ( from the business model of agent developers ) harness. Personally I have used only OpenCode / Codex / ClaudeCode / Antigravity.

Happy to hear India got mentioned by the same executive during that conversation. We are just started buying from them. I went to Semicon India 2026, held in delhi last week. The energy was amazing. Really pleasant. all big names including ASML.

Same here. We are at a stage where we are evaluating cost/benefit for my enterprise. I am heavily pushing for in house hardware with open models.

You might be surprised how many organisations/ teams now don’t even review pull requests. Claude does coding, Codex do code review, and if developer did 5 round of this before pull request, we could very well just let CI merge.

So when can we start getting cheap chips? RAM anyone please!

YouTube Channel ClockStack.

https://youtube.com/@theclockstack

4 regions analog clocks running live 24x7.

USA, Europe, India, EastAsia

Currently FullHD, upgrading to 4K if it becomes slightly popular.

Yea, I have started using c++ for some of the stuffs previously used python for. When agents is doing the grunt work, better to go even closer to machine.

At this point, I haven't even read around 30% of the code base in my open source project. I know our works by my manual testing. AI keeps writing tests for itself, even though I don't explicity ask for it, and I am not complaining.

A simple framing for reference.

Electrical Grid Frequency:

50 Hz = 3000 RPM.

60 Hz = 3600 RPM.

So, for Gaming Monitors 144 Hz = 8640 Hz.

Maybe this is one of those topics to let an army of agents get loose.

Nothing to lose and a lot to gain.

Yes. Waiting for the Windows Version myseflf. RTX Spark Desktop.

I hope it brings the RAM prices down!

Same here. Late to agentic party. 1st month was Codex, 2nd Claude, already thinking 3rd will be Codex again! Since Fable will be gone soon.

For me the Best Pros is "That one random-internet-comment is a good usability improvement, ask LLM to do the small change and commit immediately."

Smaller ideas need not be approved, or held back by schedule pressure from bean counters. Just Do It. :). It's the small corrections which end up polishing the product as good as "professional usability studies".

If they release a model comparable to OpenAI / Antropic, will there be any reason left for 1T valuation of other companies? At that point, it will simply become Revenue proportional to Gigawatts available. Whoever got the energy wins.

I can count 2:

Dec 2025: We use a commercial 3D modeling software to build refinery. There was no license dashboard in this ancient piece of junk. Fortunately license server provided verbose live status report through a command line. I ask ChatGPT to ingest the logs into a Django web application and generate weekly/monthly/yearly usage dashboard, and It one shorted the whole Backend + Frontend in 4 to 5 shot. There were around 10 regexes just in the log parsing batch script. I was totally speechless. Encouraged by the success of, I went ahead and made the dashboard for 3 more software in the same Django app. Released to peers by evening, feedback incorporated in 2 days to integrate Name, Employee Number, IP Address sync etc in 2 days. And it’s been live for 5 months, actively being used by all coadmins, even management has it bookmarked, to help with department redistribution. Making this thing without AI would have taken well over a month of “learning new stuff”, or paying external consultants too much. Even head of IT replied back, it was awesome. ;)

2nd , June 2026: I asked codex to something fairly complex before going to morning bath!, which would have taken me more than a week of learning DirectX12 API nuances and such things, 20 min latter, I return to task exactly completed with code changes in 5 different files. Build complete without any error. OMG. Free Quota over for whole month! I subscribed by the evening.

Once someone taught me "you can do xyz reading a book, but you cant do surgery by reading a book". Now replace the book with LLM. This is what "domain expertise" look like for some domain.

Agree on this one. As a Civil engineer, I can smell which software ( or some part of) was developed by computer engineers without much understating of the "domain". The worst offenders are software needing multi-domain expertise in the first place. Crude Ex. Payroll (Finance) and Leaves (HR) are 2 seperate domain, and they need to account for each other.

Guess we can run this even on CPU!

With the new agentic capabilities, I am quickly running out of Architecture decisions I have already made myself! For my work-in-progress engineering application. There is also some kind of don't know every little if/else with my own Code now.

However the good part, what I had planned for 5 years, now looks like doable in 6 months. Looking forward to real use by the end of this year.

Ref: https://github.com/ramshankerji/Vishwakarma

Thank you for the insight. This makes me feel confident, the L40S we are about to acquire with 48GB VRAM for engineering application should be useful for agentic coding as well.

I wish it offered a windows binary as well, since the original project gitea from which it was forked, does so.

Sometime you need to go where many customers ahem enterprise are.

As a developer of an engineering application, windows is the way to go 1st hand. It would have been easier to adopt one more application on the daily driver enviornment. Till than, I am on GitHub only.

This is the best publically posted model size, ever since top AI labs started treating model size as a trade secret. This should also guide next generation of inference ASICs.

Building blast resistant is a common practices for Refinery control rooms. The same methodologies can be employed for data centers as well.

1 blast can be expensively guarded againt. However designing anything above ground for sustained barges is practically/commercially prohibitive. Underground is only option.

PS: Civil Engineer. Designed few of those Gas explosion resistant control rooms.