engineer @datanimbus | Was - Chief Evangelist at #NeebalTechnologies | engineer @coherendz | can #work for {food} | runs, reads and travels
Joined June 2009
- Tweets1.5K
- Following478
- Followers188
- Likes357
This one stays in Bombay, and I am officially the "operator" for this ENIAC.
github.com/jace/minirack
Thanks @jackerhack , @zainabbawa
United by FOSS, ferried by @IndiGo6E back safe and sound to the pad by herder @jackerhack
Where will.ypu go next?
@FOSSUnited @MumbaiFOSS @ravimbalgi
Ravi Balgi retweeted
4/5
31 October · IIT Bombay, Powai
Attending in person? Get your Mumbai FOSS ticket from fossunited.org/c/mumbai/2026
Schedule: fossunited.org/dashboard/sch…
Ravi Balgi retweeted
3/5
1:30–2:30 PM IST · @yashgandhi_ shares about local analytics, a transformation language analysts won't learn, and an agent that turns plain-English requests into validated workflows, in his talk "“Big data, mighty machines”. @OrcaSheets
Ravi Balgi retweeted
2/5
10–11 AM IST · Swetha Ashok from @SahajSoftware will speak about how a production assistant’s agents arrived at conflicting answers, and how shared understanding and explicit state helped.
Ravi Balgi retweeted
1/5
Tomorrow at Mumbai FOSS: two Enterprise AI in Production talks, selected and shaped through peer review.
Curated by @ravimbalgi with support from @neosrix to bring practical experience into the programme.
Ravi Balgi retweeted
5/5
Watching online? The live stream is FREE for The Fifth Elephant annual members.
hasgeek.com/fifthelephant/en…
Thank you @ravimbalgi and @neosrix for the editorial work behind these sessions.
Come and learn to understand the work done to put AI to work within Enterprises.
Happening now at @iitbombay
through @MumbaiFOSS and @hasgeek
Hasgeek finally arrives in #amchimumbai
has.gy/hDJ_
#technology #ai #OpenSource
Ravi Balgi retweeted
Around 3 weeks ago introduced Laya, the first open-source version of JEV, and it has reached a very large scale. Today I am open-sourcing a new physics-based typed decision model called Vega.
The interesting parts. It is only 800 million parameters (4B is also there), with 73k token support, image support, and surpassing many of the jev benchmarks in single-shot. Implemented an engine+adapter as a Test-Time Training Architecture.
Imagine you are giving input to the model like "ignore all previous conversations," which is called the observation, along with a question, "Is this trying to override the instructions of the model?" and your possible outcomes are YES or NO. The whole prompt is passed to an LLM/VLM (yeah, image-supported), then from the hidden layers, we can extract the observation, question, the yes and no, and last token. These will be vectors, and we can project it by multiplying with wieghts; then, first, using the observation projection, we can create a landscape with valleys; then, using the question and last-token vectors, we can initialise a ball in the valley, and using YES or NO, we can get the candidate location of the ball. Depends on the number of outcomes we create valleys; that is, here YES and NO, so 2. When the ball fall on YES valley, we get the answer. There is friction also influencing how the ball moves.
TLDR: It's like creating valleys of outcomes that we need and throwing a ball that will slow down based on friction and settle on the best outcome
Blog: nandakishorm.com/writing/veg…
GitHub: github.com/NandhaKishorM/veg…
HuggingFace Model Card: huggingface.co/nandakishorm/…
HuggingFace Space: huggingface.co/spaces/nandak…
Ravi Balgi retweeted
Nine minutes after Adam Neumann stepped down as WeWork's CEO, his bank emailed him. He had 15 days to repay about $450 million, with interest compounding daily.
In January 2019 SoftBank had valued WeWork at $47 billion. Then the IPO filing went public. It lost $1.9 billion in 2018, and it was signing long leases on whole buildings while renting desks month to month. Investors walked. By September the IPO was dead and he was out.
His fortune was WeWork stock. He had borrowed against it, and the loans had a clause: if he lost control of the company, they were in default. Pay in 15 days, or the bank takes the shares. He had the shares and no cash.
A few nights later he woke up crying next to his wife, Rebekah. It was the first time she had ever seen him cry.
"We lost everything."
"What do you mean everything?"
"Everything."
She reminded him of the $100 million they had agreed to put in her name, just in case. He had never done it. And the personal guarantee they had agreed he would never sign. He had signed it.
She didn't blame him. "Hm, okay, this is a surprise." Her mother owned a house upstate, bought 30 years earlier, no mortgage. "Let's go live with my mom. Take a few months. If you want to go back into business, I know you'll be successful. And if you don't, we'll move to Costa Rica."
Then: "When I met you, you were broke. I think you're much more sexy when you're broke." He smiled.
Two broken promises, a $450 million hole, and she answered with a plan and a joke. That is what a partner looks like on the worst night.
SoftBank paid off the banks and took control. In 2022 Andreessen Horowitz put $350 million into his next company. His rule now: "Choose your partner for the day something like that happens, not for any other day."
Ravi Balgi retweeted
1/ Tomorrow at The Fifth Elephant 2026 Annual Conference, we have stickers for everyone building in the agent era and quietly discovering that the hard parts are still the hard parts.
Fifth Elephant 2026 is live from NIMHANS Bangalore
Streaming on
youtube.com/live/Etyrz09XbSg
youtube.com/live/ygidsbBiDVs
#fifthelephant
cc @hasgeek
Ravi Balgi retweeted
Nobody solves cross-team data sharing without duplicating everything.
@InMobi's Shivam Gupta built DataBridge on one rule: discover globally, materialize locally. Query across lakehouses in place, copy data only once it proves its worth. @shivamgupt33983
Saturdays done right with grounded folks, practitioners and friends at मुक्काम.. पुणे learning about Indian AI models in practice
Courtesy @thoughtworksIN and @hasgeek. #pune
cc @zainab, @sitaramshelke, @dzouza_craig , @kiran_kulkarni
Run up towards @fifthel in #ooru #bangalore
Prathamesh Kalamkar explaning hard lessons from training domain specific LLMs while building legal asssistant model for Aalap
@hasgeek #fifthel #pune
@kprathamesh
cc @dsouza_craig
Ravi Balgi retweeted
He cleared NEET twice with top scores but still found himself outside a classroom because the EWS reservation in Madhya Pradesh private medical colleges was never implemented.
At just 19, Atharva Chaturvedi walked into India’s Supreme Court armed not with emotion, but with the Constitution, past judgments, and his own petition. He had studied law from online court videos at home in Jabalpur, drafted his Special Leave Petition himself, and filed it online to save travel costs.
When the bench was rising for the day, he asked for ten more minutes to present his case. They listened.
In less than ten minutes, the Court invoked its extraordinary constitutional powers and directed that he be granted admission under the EWS quota. The judges noted it was law, not theatrics, that carried him through.
Atharva’s story is unforgettable because it is not just a legal victory. It is a young aspirant from an economically weaker background learning law on his own, standing before the country’s top judges, and winning through clarity, courage, and persistence.
Soon, he will step into a medical classroom that he had been denied—through justice, not chance.
#Inspiring #NEET #EWSReservation #SupremeCourt
[NEET, Supreme Court, Atharva Chaturvedi, MBBS admission]
Ravi Balgi retweeted
Final demo for the day - Sovereign LLM Inference on Heterogenous AI Accelerators Using llm-d and vLLM by Pravein Govindan Kannan of IBM Research.
has.gy/INl_ – at Anekal Bangaluru, India
Ravi Balgi retweeted
Live from today's meet-up on Enterprise AI in Production. Akash Satish is speaking about vulnerabilities involved in running MCP servers.
has.gy/zOZQ
@SahajSoftware – at Anekal Bangaluru, India
Ravi Balgi retweeted
Falling in love with someone can teach us how we can be in love with everyone and everything.
When we are in love with someone, everything about them seems indescribably perfect.
But when we look deeply into what is happening there, we can see that the love that reveals their perfection is not in them but in us.
Our love is like a light we are shining on them, illuminating their lovely qualities.
But it is also the loveliness of the light itself that we are enjoying.
It is a great meditation to do when we are in love:
To meditate on the quality of the love, independent of the beloved.
To familiarize ourselves with it and see that it radiates from within us.
When we see that this light is shining from us, we can learn to shine it on anything.
In doing so we can learn that we do not love things because they are lovely.
But rather, what we love becomes lovely.
I do believe this is a large part of what we are here in this universe to do.
To learn how to recognize that light.
To learn that on the deepest level we are that light.
It is not our lover that makes us whole when we fall in love.
But rather it is the love that we love them with that makes us feel whole.
Because in it we catch a glimpse of our true nature, which is a love that is already whole.
