backing AI-native founders from first cheque to IPO.
Everywhere
- Tweets 312
- Following 83
- Followers 1,516
- Likes 2,839
Pinned Tweet
meet boundless ventures.
AI-native global seed capital. from india, for the world.
boundlessvc.com
8
3
52
12,766
india's quest is to build a global space economy. the talent and policy movement are here, can the capital keep up?
our founder, @natashamalpani was featured by @ETtech on why our portfolio company @piersightspace is a global outlier. their data assets will only become more irreplaceable. on that note: if you're a space-tech engineer, we're hosting a space design workshop this friday in ashok nagar, bangalore. registration link for zero to orbit: a spacetech founder simulation is in the comments. article link also in comments.
@SmruthiNadig
2
2
17
533
registration link: luma.com/xsq16kyb
1
4
53
article link: entrepreneur.economictimes.i…
1
5
127
article link: entrepreneur.economictimes.i…
registration link: luma.com/xsq16kyb
2
19
most spacetech engineers know their function better than almost anyone in the world.
very few get to think across product, capital, team, and first customer at the same time under pressure, from scratch.
zero to orbit reverses that for one evening.
pods of 4-5 engineers and researchers pick a theme orbital servicing, space manufacturing, orbital data centres, relaunch infra and have 45 minutes to build a globally competitive startup. then they pitch.
curated on the 18th of september in bangalore. link in comments.
1
12
461
registration link: luma.com/xsq16kyb
1
5
290
registration link: luma.com/xsq16kyb
1
15
cost per minute for voice AI has gone from a few cents to fractions of a cent in about a year. for enterprise and prosumer tools, quality comes first early on, and customers are willing to run expensive frontier models to get there. cost only becomes a live concern once the product has found its footing.
over the last couple of weeks we've had some genuinely interesting conversations on inference. we put some of what we found into writing.
link in comments.
1
2
7
354
we hosted a highly curated group of inference engineers and researchers at @lossfunk running through what actually breaks once a model hits production traffic.
some fascinating takeaways:
1) the metric a team optimises for tracks their business stage, not their engineering instincts. before PMF, quality and TTFT win, because the whole point is proving the experience works. after, cost takes over, because a team's customers have margins of their own to protect.
2) latency and cost fail in completely different ways. a slow voice bot breaks the call outright; the caller talks over it, and the interaction falls apart. a slightly expensive content pipeline breaks nothing; it just eats margin until volumes increase bill amounts significantly.
3) india isn't a cheaper version of the same problem. voice buyers here won't go much past two rupees a minute. in the US, five to seven cents is the average. at the indian price point there's no margin left to buy a good asr, a good llm, a good tts, and telephony from four separate vendors. so indian voice teams end up owning the whole stack, down to writing their own sip layer. almost nobody in the room had chosen their current stack the way a case study implies. they'd arrived at it, tossing away one constraint at a time. the final stack wasn't the best one on paper. it was the one that didn't get vetoed.
4) for anything regulated, or running on a factory floor with patchy connectivity, none of the above matters. the cheapest option on a spec sheet is irrelevant if it can't guarantee where the data goes or whether the connection holds.
thank you to everyone who brought their technical opinion, some spectacular, very specific insights.
thank you @lossfunk, we appreciate the space you hold for us and the space you allow us to use.
1
2
18
846
proud to support dreamers who rethink systems from first principles
@khushhhi_is reimagining how we move goods across the world
For generations, we've pushed the limits of what flight can do.
Yet moving goods still comes with a compromise. The sea takes weeks and flying costs too much. Somewhere along the way, the world got comfortable with waiting.
Today, we announce Eureka, our first aircraft. It takes off and lands on water, and it's designed to turn "in a few weeks" into "on the evening flight."
Break the surface.
1
1
16
1,182
we're getting ten to twelve inference engineers and researchers in a room on august 29th, with our friends at @lossfunk.
some care most about speed, some about cost, some are somewhere in between. we want to see what happens when people optimising for different things sit at the same table and build a shared way of thinking about inference together.
we're calling it safe to infer: open, closed, and what's actually in production.
it comes off the back of something we wrote recently. most AI products start the same way: you just pay per use through an API, no infrastructure needed, fastest way to get going, and for most of them, that's exactly where they should stay.
but a few things push some products down a different road. when speed becomes non-negotiable. when you need to prove exactly where your data went, or when you're doing so much volume that paying per use stops making financial sense.
we wrote about what that road actually looks like, and where the value ends up once companies go down it.
read it, and register for august 29th, both in comments.
1
1
9
280
registration link: luma.com/3euc26s9
link to article: boundlessvc.substack.com/p/t…
2
97
proud moment for our portfolio @piersightspace 🙏
Met the Hon’ble Prime Minister Shri @narendramodi ji for the first time today, and had the opportunity to share what we’ve been building at @piersightspace.
To my surprise, he already knew what SAR technology is and how big of a role it will play in building India’s sovereign space capabilities.
When the PM affirms the demand for the technology we’re building, what more could I ask for as a founder?
All in all, I had an amazing conversation & memory for a lifetime, ending on a funny remark that he felt I was looking more of a politician than a scientist! Well, now I feel I did look that way :)
Thanks to @INSPACeIND @GoenkaPk @PMOIndia for inviting me to this interaction as part of the National Space Day Celebrations.
#NSpD2026
1
11
436
sixteen hours after our AI salon in bangalore, we took it to delhi.
The same central question on the table: "what did you believe about AI a year ago that you don't believe anymore?" five people to a pod, working it through together, then everyone came back to compare notes, and then we ran an open challenge clinic. people brought whatever problem they were actually stuck on and worked it out with the community.
we had founders building across healthcare, fintech, defence tech, and consumer AI, with some AI engineers and researchers in the mix.
a big thank you to @AICollectiveCo, @themastersunion and @callmeshuklaji
1
16
873
a voice engineer worries about time-to-first-token. a content engineer worries about cost-per-output. put different contexts at the same table and see where they diverge, converge, and what we can infer.
we're putting together a small technical afternoon for researchers in the space and engineers running inference systems day to day, across different optimisation regimes, latency-obsessed, cost-obsessed, mixed tradeoffs, plus infra people who watch the pattern repeat across all of them.
ten to twelve people, saturday, august 29th with our friends at @lossfunk. we start from different priority metrics and build a shared framework for thinking about inference from wherever the insights lead.
deep in inference and want in? link in the comments.
1
8
1,615
we hosted an AI salon with @amuldotexe and @lossfunk built around a single question.:
what did you believe about AI a year ago that you don't believe anymore?
over 400 people applied to working theories.
we split the carefully curated teams into five-person pods to go deeper. then everyone came back together and shared what their group had landed on. we closed with an open floor. people brought a problem they were actually stuck on, and the room worked through it together.
a few things came up again and again through the afternoon. whether verification matters more than generation now that code writes itself. whether agent evals are quietly becoming the new unit tests. whether judgement has overtaken output as the scarce skill.
thank you to lossfunk for the space and to amul, who steered the room.
stay tuned for the next one.
1
19
72
9,839