We’re reimagining how enterprises process documents. Not with another tool, but with an AI layer that understands context, structure, and intent. #BuildInPublic #AI #KnowledgeVerse Senthil Vijay Nadadur

We’re reimagining how enterprises process documents. Not with another tool, but with an AI layer that understands context, structure, and intent. #BuildInPublic #AI #KnowledgeVerse Senthil Vijay Nadadur delta

Each step seems easy, as long as you observe it from a distance. The challenges are realized only when get your hands dirty. #BuildInPublic #AI #KnowledgeVerse delta

This week, our messaging and design sprint has been underway: rewriting our one-liner and 50-word pitch. Sometimes, the hardest part is explaining something powerful, in simple words. #AI #KnowledgeVerse #buildinginpublic Vijay Nadadur Senthil delta

This week, we were committed to cover the essentials for startup! We started with new decks for investors and customers. Each slide answers one question: What? Why, Why now? Why us? #StartupJourney #AI #KnowlegeVerse #buildinginpublic delta

Our one-liner that sums up what we are doing at #KnowledgeVerse "The AI Layer that bridges gap between document workflows and LLM". #BuildInPublic #AI delta Senthil Vijay Nadadur

Most of the tasks that were planned for last week are done. This week, implementing the latest design, updating the website, and initiating customer outreach are our goals. #buildinginpublic delta KnowledgeVerse AI

Excited to share the we are launching our new website tomorrow. Watch this space. delta KnowledgeVerse AI #buildinginpublic #BuildinPublic

The thing with initial days of a startup is: even if you take a break for a day, you start to feel guilty. Especially with the speed at which everyone's releasing new products and features, seems like the race is always on. #AI #KnowledgeVerse #BuildInPublic delta

If you are in the business of Document Processing, you can't ignore #DeepSeek's new OCR offering. alphaxiv.org/models/deepsee… (1/2)

#DeepSeek-OCR is a unified end-to-end Vision-Language Model designed to address the long-context problem in Large Language Models through "contexts optical compression." #NLP #ComputerVision (2/2)

This was super productive. So, what have done? 1) Launched our new website. k-v.ai 2) Launched our upgraded platform: platform.k-v.ai 3) Spoke with several potential customers. #BuildinPublic delta @KVerse

Spent more than 40 hours speaking to various industry leaders and other decision makers. Gave them a walk through of the latest platform. Got 20+ signups. (3/n) #announcement #SalesProcess #B2BSales #KnowledgeVerse #BuildingInPublic delta

This week's highlights: 1) Website released 2) Updated Platform released. 3) Sales in motion with over 20+ signups. (4/4) #announcement #newwebsite #KnowledgeVerse #BuildingInPublic delta

"Doubt is not a pleasant condition, but certainty is absurd." - Voltaire Founders, who have never worked in enterprise are absolutely certain about the problems faced by enterprises. They exactly know how agentic AI can fix it all, and are certain about every aspect of

New research drop from KnowledgeVerse AI — 3 papers on why LLMs still struggle with self-awareness & reliable reasoning. 1) Line of Duty (aclanthology.org/2025.trustnlp-…) LLMs contradict their own “I can / I can't” boundaries ~20% of the time. Self-knowledge isn’t solved; models don’t know

2) Mirage of Mastery (arxiv.org/pdf/2506.18998) Models confuse memorization with reasoning. Slightly perturb a problem → confidence collapses. Benchmarks ≠ real intelligence. 3) TeXpert Benchmark (arxiv.org/pdf/2506.16990) First dataset for natural-language → LaTeX generation.

Takeaway: LLMs are powerful, but raw models are unreliable for enterprise workflows. The future = AI middleware that adds memory, validation, context management & trust. We’re building that at KnowledgeVerse AI (3/3)

In the nutshell, this is how data pipelines look like, when you want to leverage LLMs. It's simple than you think, yet more powerful than you realize. [1/2]

And this is how KnowledgeVerse AI further helps you improve your pipelines tremendously. Remember, it's not the model / OCR / RAG problem, but core engineering problem; a combination of data and context, problem. Improve these, and see results improve drastically. [2/2] #KnowledgeVerse