Pramod Shri

145 posts

Pramod Shri banner
Pramod Shri

Pramod Shri

@XPramodShri

Founder & CEO at Knidal. Builder. Learner.

Everywhere Katılım Mart 2010
218 Takip Edilen88 Takipçiler
Sabitlenmiş Tweet
Pramod Shri
Pramod Shri@XPramodShri·
OpenClaw is now published on Knidal. > Humans get a great reading experience at: openclaw.knidal.com > AI systems can access the same knowledge at: api.knidal.com, using the slug openclaw "AI-native content. Human-first experience."
English
1
2
5
85
Pramod Shri
Pramod Shri@XPramodShri·
We need to make the internet AI-ready. That means publishing knowledge so AI systems can retrieve it cleanly, while the human experience stays first-class. That’s what we’re building with Knidal. If you want to publish knowledge (docs, manuals, guides, etc.,) that are great for humans and usable by AI systems, DM me.
English
0
0
0
14
Pramod Shri
Pramod Shri@XPramodShri·
Compared to pulling from a traditional docs site, this is a much better setup for AI retrieval. Instead of working through presentation markup, indirect navigation, and noisier extraction paths, AI systems can work with a cleaner, more structured knowledge interface.
English
1
0
0
13
Pramod Shri
Pramod Shri@XPramodShri·
OpenClaw is now published on Knidal. > Humans get a great reading experience at: openclaw.knidal.com > AI systems can access the same knowledge at: api.knidal.com, using the slug openclaw "AI-native content. Human-first experience."
English
1
2
5
85
Niels Rogge
Niels Rogge@NielsRogge·
New blog post: converting 30k @arxiv papers to Markdown using SOTA OCR models to enable chat with paper functionality Includes: > leveraging an open OCR model (Chandra 2 by @datalabto) > running on GPU infra - @huggingface Jobs > using Codex with a SKILL.md
Niels Rogge tweet media
English
12
70
541
54.5K
Pramod Shri
Pramod Shri@XPramodShri·
If you're hosting a knowledge base, docs site, product manuals, guides, or a help center, and you want content that's ready for AI and pleasant for humans, let's talk. knidal.com
English
0
0
0
22
Pramod Shri
Pramod Shri@XPramodShri·
This is what we're building at Knidal: AI-ready content with a human-first experience. Always available like a book. Always current like software. Rich with media, searchable by meaning, and ready to answer questions instead of just sitting there.
English
1
0
0
25
Pramod Shri
Pramod Shri@XPramodShri·
Every PDF converter is terrible for anything beyond the simplest documents. @karpathy just hit this with a 245-page PDF and every tool fails. This isn't a tooling problem. It's a format problem. And the fix isn't a better parser. 🧵
Andrej Karpathy@karpathy

@kepano I just tried it this morning on the 245-page Mythos pdf and it failed badly and the outputs were all mangled. Converting pdfs is really hard, I think it has to probably be a Skill not a program, for a SOTA LLM for it to work properly.

English
1
0
0
32
Pramod Shri
Pramod Shri@XPramodShri·
@burkeholland Markdown is not the full answer. But it is still a better starting point. It is clean, structured, token efficient and low on noise. Then you add deterministic checks, validations on top.
English
0
0
0
6
Burke Holland
Burke Holland@burkeholland·
Markdown is not the answer to your AI problems. It just can't be. We need to force AI through deterministic gates.
English
104
13
258
41.8K
Pramod Shri
Pramod Shri@XPramodShri·
Also, one time conversion is not enough for living knowledge. Content changes all the time. If one new chapter means reingesting everything, the workflow becomes expensive very quickly.
English
0
0
0
11
Pramod Shri
Pramod Shri@XPramodShri·
We have seen this firsthand in our platform. When content is organized in an AI ready way instead of being passed through as raw documents, token consumption drops a lot and response quality improves meaningfully.
Andrej Karpathy@karpathy

Use epub not PDF, convert epub to txt/md, summarize the wikipedia article into "book context", with it in context summarize one chapter at a time, etc. I mean basically imo for good results you have to "work it" in chunks and shouldn't expect that just attaching a pdf and asking summarize will give good results. When I do it in stages and slower I can get very good results, indispensable.

English
1
0
0
16

Keşfet

@NielsRogge @arxiv @datalabto @huggingface @karpathy @burkeholland @elonmusk @BarackObama