Nagajyothi

Building AI products full-time, in public.
Currently: Dashlytics. ( for build logs, not startup advice

[Nagajyothi @njpCoder]

Update #2: The Adversarial Buyer is taking shape.

The goal sounds simple:

Find the objections that can actually kill a deal.

But I don’t want an LLM to just look at a landing page and hallucinate 10 generic complaints.

So I’m making it earn every objection.

→ Crawl what the buyer can actually see
→ Extract every claim the company makes
→ Attack those claims from different buyer personas
→ Remove objections the website already answers
→ Cross-check high-stakes claims against external evidence
→ Rank what survives by potential damage
→ Trace every objection back to the exact line that caused it
The output: ranked objections, with receipts.
Architecture locked.
Now building it. 👇

[Nagajyothi @njpCoder]

Build 2 of 3, done.

A dental clinic where requests arrive by WhatsApp and phone, someone retypes them into Excel, five dentists share three chairs, and follow-ups live nowhere at all.

Chairside replaces the retyping. Not the receptionist.

[Nagajyothi @njpCoder]

Runs local. Their repo, their config, their tokens. It's opensource. Comment "punchcard" if you want to try it.

2pm slot next: automating appointment booking for a dentist.

[Nagajyothi @njpCoder]

Build 1 of 3, done.

The worst 20 minutes: filling in a timesheet on Friday, from memory, for a week nobody remembers.

punchcard reads the calendar and the git history, groups it by day, and asks you to map each source to a project once. After that it remembers.

[Nagajyothi @njpCoder]

The decision that actually matters: it surfaces 21h 45m as unaccounted and refuses to fill it.

Every auto-timesheet tool pads to 40 hours. That's the feature that gets you audited. If punchcard doesn't know where an hour went, it says so and leaves the call to you.

Dry run before anything is pushed. Jira worklogs, Toggl or Clockify on the other end.

[Nagajyothi @njpCoder]

@pebb_io It's opensource. Try it out of you are interested.

https://t.co/mScF3xTdVi

[Nagajyothi @njpCoder]

Three people sent us the most boring 20 minutes of their week. We're building all three away today, 9am to 9pm.

10am - Automate timesheet submission for devs
2pm - Build a content engine for a new instagram page
6pm - Build appointment booking system for a dental clinic
Bonus - SEO audit for a clinic website

They keep everything. We keep nothing, charge nothing, maintain nothing.

[Nagajyothi @njpCoder]

We are posting the challenge on instagram: https://t.co/YP1jcr626i

[Nagajyothi @njpCoder]

The rule we're testing: don't take feature requests, take a 20-second screen recording of the manual version. The recording is the spec, and it's the part nobody sends you.

The ones that break get posted too.

[Nagajyothi @njpCoder]

Nobody describes their problem as a problem.

They describe a Friday. Export the file, delete four columns, paste it into an email. Forty times. To them that's not a broken workflow, that's just the job.

Which is why feature requests are worthless and screen recordings aren't. The manual version is the spec.

Tomorrow 9am–9pm we're taking 3 of these live. Send the worst 20 minutes of your week, we build it away, you keep it, free.

Reply with the boring thing you do every week.

[Nagajyothi @njpCoder]

Websites are starting to expose tools directly to AI agents - WebMCP.
OpenAI shipped support last month, with Shopify and Cloudflare as launch partners.

553 sites and 3,648 tools are live right now. Almost nobody is checking whether those tools are safe to hand an agent.

Built a free scanner. Paste a URL, see what agents see:
https://t.co/COM4N1R5Dh

Open source, MIT. Reads the contract, never calls a tool.

[Nagajyothi @njpCoder]

@namanambavi @OximyHQ Adoption of AI within organizations is rapidly becoming complicated. The visibility layer may prove to be as important as the AI tools themselves.

[Nagajyothi @njpCoder]

@agazdecki This is what makes bootstrapped SaaS so fascinating to me. It doesn’t require you to have a massive team or millions of dollars invested in it. Just identify a painful enough niche and concentrate on your product.

[Nagajyothi @njpCoder]

@Ozacle23 Looks so cute

[Nagajyothi @njpCoder]

First time submitting to @hackernews today.

I went in thinking it was another place to post your product.

Nope 😂

HN has its own culture, etiquette, ranking, Show HN rules… even the way you write your title matters.

Still figuring it out, but I kind of like that it forces you to drop the marketing language and just explain:

What did you build, why did you build it, and why should anyone care?

Hacker News veterans - what did you wish you knew before your first submission?

[Nagajyothi @njpCoder]

@gregisenberg Love the video

[Nagajyothi @njpCoder]

@kaiskhimji Sometimes the most interesting thing on a market map is the company that isn’t on it yet.

Themes can be right long before the winner exists.

[Nagajyothi @njpCoder]

@ThePeterMick Yesssss😂

[Nagajyothi @njpCoder]

@readswithravi Big goals, tiny moves, relentless consistency.

[Nagajyothi @njpCoder]

Agent harnesses are having their moment.

For a long time, they were treated as “just scaffolding” around the model.

But memory, tools, context management, sub-agents, persistent state - all of that determines how much of the model’s capability you can actually unlock.

This @ycombinator Paper Club on why the harness matters is worth watching 👇

https://t.co/7ym7DhY2ml

[Nagajyothi @njpCoder]

Bengaluru folks who’ve house-hunted recently - I need your help 👀

I'm building ನೆಲೆ - Nele 🏠

A way to find homes based on how you actually live - not just BHK, budget & location.

If you’ve ever thought “I wish the listing told me THIS before I travelled all the way there” - this is for you.

Join the waitlist → https://t.co/yrF6wWUooJ

And tell me: what’s the one thing you wish rental listings showed?