Here to learn, unlearn and relearn | Building AI products - Agentic web search, workflows and decision intelligence
India
- Tweets 5,670
- Following 445
- Followers 542
- Likes 1,854
Pinned Tweet
Being self aware helps you to own consequences.
Ownership of consequences leads to courage.
Courage brings mental clarity.
Mental clarity leads to conviction.
And, conviction leads to decisive action.
705
DeepSeek entering agentic search is another sign that this space is moving fast. The competition is no longer just about who has the best search results. It’s about who can help AI agents find information, navigate sources, gather evidence, and turn it into something useful.
1
13
I'm surprised by what all Manus 2.0 can do... that too at good speed and explainable interf.
Thanks @AbhiSuryawanshi for re-introducing me to Manus.
1
1
43
Pronunciation matters. The difference between I concede, I concealed, and I conceived is the exact difference between losing an argument, confessing to a cover-up, and accidentally announcing a pregnancy.
Enunciate responsibly.
22
Just started working with Image and video gen AI
Suggest me best models to try out beyond GPT and Argon (Google Veo)
26
Built a small poc to solve a real business problem with DeepSeek v4.1 and Nemotron (NVIDIA) on OpenCode
It worked well. The plan was prepared using Codex, execution on OpenCode.
1
39
Eat 5 star & do nothing 😉 Indian stock market is becoming the biggest, most liquid place for global investors to park money as a hedge when they sell or reduce crowded AI bets.India's growth comes from domestic drivers (consumption, banks, infra), not AI hype.
24
We spent the last three years amazed by what AI knows. We’re going to spend the next ten amazed by how it thinks. Information was just the training wheels. Inference, reasoning on the fly over messy, brand-new context is the actual moat.
21
Saurabh Jambure retweeted
Why would I pay $15/month for your software when I can plan, build, review, improve, host, launch, scale, secure, patch, monitor and debug it myself?
470
383
5,852
215,671
The most interesting AI systems may not be the ones that gives the best answer today.
They may be the ones that gets better because they remember choices, mistakes and outcomes.
Intelligence that compounds is a very different product.
24
Agentic AI Day was full of insightful sessions and good learning at @coditasHQ by @AgenticAIFdn
Cherry on the cake - Ranked 3rd in the quiz amongst 100+ participants ☺️
Looking forward to more events like this that bring together and boost the AI community in Pune.
1
29
Looking at layoffs at big cos and uncertainty (of Claude/gpt releasing their core as feature) at early stage startups - what's a safer choice?
Startups with PMF?
37
Even if the talent comes back or stays In India, I wish they CREATE more and don't just become CONSUMER.
Indians have made a major contribution to America's technological prowess. Yet, a big part of America no longer wants us.
Indians who are on a visa in the US face the painful prospect of uprooting the life you have built with hard work.
Let this be the birth pains of a bold new future.
Come back, let's build India 🙏
26
काही कर्जं पैशांनी फिटत नाहीत; ती आयुष्य देऊन फेडावी लागतात. आणि काही ठिकाणं अशी असतात, जिथे फेडलेली कर्जं सावली होऊन उभी राहतात.
नवीन लघुकथा...
#Marathi #मराठी #Story
marathi.pratilipi.com/story/…
28
I’m increasingly interested in AI that lets you explore a decision before making it.
Not 'predict the future' but more like: If we choose this, what could happen?
What are we assuming?
What could make this decision wrong?
22
Document extraction looks solved until you see 40 different layouts of the same document. The AI model isn't the hard part. Variety is. Build for the messy 20% first, and the clean 80% takes care of itself.
30
Most agent demos fail in production for one boring reason: nobody defined what 'correct' looks like.
Before you pick a model or a framework, write 50 examples of right and wrong answers. That's your eval set. Everything else is guessing!!!
25
Imagine an AI that doesn't just help with today's decision, but remembers the assumptions behind it, tracks what changes and learns from the outcome. Every decision becomes context for the next one. That's a different proposition from asking AI the same questions from scratch.
24
Many of us talk (voice) with the AI today. The real time human like interaction will enable us with DIY repairs,
Physical therapy or form-correction exercises.
Technical training, Sports technique or instrument practice feedback and much more! Awesome @hassaanraza
Introducing Griffin, the first model to pass the video Turing test.
48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.
It’s the first Human Interaction Model (HIM).
Community note
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard.
cellcog.ai/blog/tavus-gri…
research.nvidia.com/labs/amri/proj…
tech-ish.com/2026/10/02/tav
1
40