Guhan

Building an automated car wash + legal-tech AI from Coimbatore. LLB from England. Ships products, not pitch decks.

Proptech -

[Guhan @guhan729]

I'm 26. I have a law degree from England.
I run an automated car wash in Coimbatore and I'm building AI software for property lawyers.
No, those aren't three different people. Thread on how this happened:

[Jensen Huang @JensenHuang]

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.

AI will transform every industry, power every company, and be built by every country.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

The world needs both frontier closed models and frontier open models.

https://t.co/AUKzoQ5Ikb

[Guhan @guhan729]

4/Small change on paper. Easy one to skip, because “just read the text” looks like it’s working right up until a table proves it wasn’t.
If you’ve dealt with structured documents — tables, forms, ledgers — has “the OCR looked fine” ever turned out to mean “the OCR looked fine at the sentence level, not the structure level” for you too?

[Guhan @guhan729]

1/A table isn’t a sentence. I treated it like one for longer than I should have.
Most of the property documents PropFiled reads aren’t paragraphs. They’re tables — survey numbers, extents, owner names — lined up in rows and columns on a Patta or Chitta. #BuildInPublic #LegalTech #PropTech #AI

[Guhan @guhan729]

3/Flatten it and a number can end up attached to the wrong row without anything about the text itself looking wrong.
The fix wasn’t a smarter model. It was asking for the table back as rows and columns instead of a paragraph — so the relationship between a cell and the row it belongs to survives the read, not just the characters in it.

[Guhan @guhan729]

@sara_nambiar @cursor_ai Woah

[Guhan @guhan729]

5/ Obvious to people who’ve shipped ML in production for years. I’m learning it as a lawyer who picked up building. Writing it down messily now rather than pretending I always knew.

Where has the “cheap” option turned out expensive in your AI loop?

[Guhan @guhan729]

1/ The “cheaper” OCR model isn’t always cheaper.

Spent part of this week testing OCR options on old Tamil property documents — the faded, dense, mixed handwriting + print kind.

#BuildInPublic #LegalTech #PropTech #AI

[Guhan @guhan729]

4/ Real question: which model is cheapest once you price in how often a person has to clean up after it?

Framed that way, the choice stopped being close.

[Guhan @guhan729]

@GohilHardy I pay cursor 20 dollar plan and Claude 100 dollar plan, each serves different purpose

[Guhan @guhan729]

@sara_nambiar Hi, What does it do? Been following your work for sometime now.

[Guhan @guhan729]

I am using Hermes managed agent from Hostinger, I am non tech person so did not know how to self host it, would love to try though. I am a lawyer and I'm building legal tech, so I use Hermes as PA to give me regular updates on the acts, my other work which needs set up crown jobs and etc.

[Guhan @guhan729]

@shiri_shh It's all about about the customers. They have a huge base and revenue

[Guhan @guhan729]

@Sathibuilds @DanielSmidstrup What is this brother?

[Guhan @guhan729]

@Sathibuilds Woah, that’s great. Building in public actually helps

[Guhan @guhan729]

@grok @HermesAgentTips I’m new to this, still exploring on what to do ? I am a non tech person

[Guhan @guhan729]

@ClaudeDevs 😂😂crazy