HelixaPre-seed
Protein design models for enzyme engineering.
Back the research
We invest before the market sees it, when the technical insight is still the main asset and before the commercial story is fully written.
[B12] Venture capital · Singapore
The best AI companies start in labs, not boardrooms
We write first cheques at pre-seed and seed for technical founders in machine learning, AI for science and developer infrastructure.
01 / Thesis
We favour deep technical advantage over early revenue. A paper, a benchmark or a working prototype often says more than a pitch deck.
→ 01
→ 02
Write the first cheque
Pre-seed and seed. We are often the first institutional money in, and we're comfortable leading the round.
→ 03
Stay close after
We work with founders on the hardest early transitions: from researcher to CEO, from repository to product, from lab to buyer.
02 / Focus
Four places we look for founders.
Narrow by design. We go deep in the areas where technical depth makes the biggest difference to how a company is built.
[ML]
Machine learning
New model architectures, training methods, evaluation and inference: the foundations that other AI products are built on.
[SCI]
AI for science
Biology, chemistry, materials, climate and physics, where models compress years of experiments into weeks.
[INFRA]
Developer infrastructure
Compilers, runtimes, data systems and tooling that engineers adopt bottom-up and then can't work without.
[UNI]
University spinouts
PhDs, postdocs and faculty taking work out of the lab. We help with licensing, structure and the move out of academia.
03 / How we help
Research into a product, a first hire and a clear market.
- 01
Product
We help you find the smallest version of the research that someone will use, and ship it.
- 02
First hire
We help you find and close the first engineer or operator: usually the most important hire you'll make.
- 03
Clear market
We help you name a buyer, a budget and a reason to act now, before the seed money runs out.
04 / Portfolio
Built from research.
All companies →
KernelworksSeed
A compiler that auto-tunes ML kernels across accelerators.
Lattice LabsSeed
Generative models for battery materials discovery.
05 / Signal
What we look for in a first meeting.
Revenue isn't required. What we want to see is that you know something important that others don't, and that you can build it.
# b12 --first-meeting founder.technical_depth // required founder.shipped_something // paper, repo, system insight.non_obvious // why now, why you advantage.hard_to_copy // data, method, talent market.plausible_buyer // we'll help sharpen it // optional at pre-seed revenue = null polished_deck = null
06 / Pitch
Building from a lab?
Talk to us early.
pitch@b12vc.com →A short note, a link to your paper or repo, and what you want to build. That's enough to start.