Pratik

2X founder.
I tweet about startups, software, and AI.
Building in public w/ @notsam____ .
Football Fan

[Richard Blythman @richardblythman]

⚡ We made a library + platform to make it easy to convert and deploy existing agent projects as MCP servers.

🧵👇

[Pratik @indiebuilderx]

MVP doesn't mean pushing a broken product, it means pushing a focused product.

[Pratik @indiebuilderx]

One one hand these corporations say that the entire workforce will get replaced by AI, and on the other hand they spend so much building massive campuses. Strange!!

[Indian Tech & Infra @IndianTechGuide]

🚨 Microsoft is building its largest R&D campus after its Redmond headquarters in Noida, Uttar Pradesh.

[Pratik @indiebuilderx]

The smartest of LLMs suck a big time on legal documents. Never use them for any important legal document.

[Pratik @indiebuilderx]

@perplexity_ai why does search keeps getting enabled every time open a new chat? It was perfectly fine earlier but from the last few days I see the error.

Not sure it is by design but I just want to use the models not the search and its so frustrating to turn it off every time.

[Pratik @indiebuilderx]

With AI taking over almost all the tasks with require writing, it seems that there will be so less of original human-generated content in the next few years to further train LLMs on which would be the next challenge for corporations to solve.

[Pratik @indiebuilderx]

Just noticed a year back I used to prompt like a Google search and now I google like a prompt.

Is anyone else seeing similar behavior changes?

[Pratik @indiebuilderx]

Please build a signal/noise classifier for my X timeline

[Richard Blythman @richardblythman]

Today is a big moment. I'm very excited to announce our launch (along with fully open source release of the code) of v1 of the Naptha Stack. This is something we’ve been building towards for years.

AGI won't be a monolith. We believe it will be a network of intelligent agents with heterogeneous models, architectures and data that adapt, compete and cooperate across many use cases.

However, it's difficult to build such systems today because:
1. Agent and multi-agent frameworks built by different groups don’t interoperate. There are few standards for agent interactions.
2. Open Source Models use a variety of standards: Format for tool calling and reasoning, and adherence to structured output, vary depending on the model.
3. Scaling multi-agent orchestration across a network of nodes is an unsolved problem.

The Naptha Stack solves this. Naptha is a framework and infrastructure for developing and running multi-agent systems at scale with heterogeneous models, architectures and data. Agents and other modules can run on separate devices, while still interacting over the network.

This is a new type of platform that will enable new types of multi-agent applications, and we can't wait to see what AI engineers and researchers build. Come help us is to reimagine the internet. Let's bring forth the Web of Agents!

Naptha Node: https://t.co/wRjjW5MRVB
Naptha SDK: https://t.co/Wn42Lz6uRF
Docs: https://t.co/5BiqtcFTn5

[Pratik @indiebuilderx]

#Hiring: Remote Frontend Dev Intern (Paid) - India

Required:
- NextJS/React proficiency
- Figma skills & design sense

Bonus: Python knowledge
Send portfolio via DM

[Pratik @indiebuilderx]

@upen946 Revenue or Profit?

[Pratik @indiebuilderx]

Learning with LLMs is so much fun and easy. The only issue is everything needs to be cross-verified, as they can make up stuff that seems true.

[Pratik @indiebuilderx]

@YesKhush_5 Just curious, did you consider selling your agency?

[Pratik @indiebuilderx]

As AI starts writing more code there would be a rise in demand for people/tools in QA and code maintainability.

AI making up logic and introducing hard-to-catch semantic errors is a big problem with all existing LLMs.

[Pratik @indiebuilderx]

@aviralbhat SEBI should also figure out how to reverse face match and delete their entire content from the internet. It is so annoying tbh!

[Pratik @indiebuilderx]

@pramodk73 Exactly. My point is that the context window doesn't matter that much here. Maybe graph RAG could work, even with that 100% automation is difficult especially on larger codebases.

[Pratik @indiebuilderx]

@Krunal778 Hey Krunal, it would be interesting to connect. I worked on a healthcare startup building AI solutions for doctors and patients. It would be interesting to chat more about it.

[Pratik @indiebuilderx]

@manoj_ahi The same happened to me lol.

[Pratik @indiebuilderx]

Using the JSON field in PostgreSQL feels like a crime.