Ventures

Where research crosses the threshold.

A venture is what a line of research becomes when it is ready to meet the world. Each one is co-owned under the compact: sixty percent to the founding fellow(s), forty percent to the commons.

AI & Machine Learning
Pre-revenueIncubating

सिद्ध (Siddha)

Instant AI model deployment powered by Reverse Synthetic Neural Networks. Synthesizes production-ready classifiers and transformers directly from client data in seconds — no GPU training, no iteration, no cloud dependency. Built for edge, embedded, and resource-constrained environments where training is impossible but intelligence is essential.

Visit site ↗
Since 202660% fellows · 40% commons
Stochastic Computing
Pre-revenueIncubating

तरंग (Taraṅga)

Development of the stochastic GPU — a probabilistic accelerator that computes over streams of random bits instead of exact arithmetic. तरंग (Tarang — Sanskrit for “wave/flux”) trades deterministic precision for radical energy efficiency, targeting ultra-low-power AI inference at the edge and in neuromorphic systems.

Visit site ↗
Since 202660% fellows · 40% commons
Market Computation
Pre-revenueIncubating

समाधान (Samādhān)

The market as a computer. समाधान (Samādhān — Sanskrit for "solution," and for the settledness of equilibrium) is built on the Market Computation Model: you encode a mathematics or optimization problem as structured orders, and the market's mean-reversion relaxes to the equilibrium that is your answer — while rival arbitrage bots accelerate the computation for profit. Solving the problem and earning the return become the same act, verified on live exchanges across SAT, optimization, eigenvalue, ODE, and shortest-path solvers.

Visit site ↗
Since 202660% fellows · 40% commons
Blockchain Technology
Pre-revenueIncubating

ठप्पा (Ṭhappā)

Mint a proof of your intellectual property — of any kind — on-chain. ठप्पा (Thappa — Hindi for the “stamp” or “seal”) lets a creator stamp a paper, dataset, design, model, line of code, or even an unpublished idea, producing a permanent, independently verifiable proof of authorship and time. Built on the lab's transformation-invariant provenance, a Thappa holds even after a work is paraphrased, reformatted, or remixed by AI — and its zero-knowledge proofs let you establish priority without revealing the work until you choose to.

Visit site ↗
Since 202660% fellows · 40% commons
Astronomy
Pre-revenueIncubating

व्योम (Vyom)

An independent verdict engine for exoplanet biosignature claims. व्योम (Vyom — Sanskrit for the open sky and the aether that fills it) is built on ExoAtlas: one ML-accelerated, pre-registered Bayesian retrieval pipeline applied uniformly to the public JWST atmosphere archive, returning calibrated posteriors and abiotic-null tests in seconds rather than the weeks a bespoke retrieval takes. Where every contested claim today — K2-18 b's dimethyl sulfide, TOI-270 d's chemistry — is argued team-by-team with mismatched methods, Vyom offers a single reproducible standard to confirm, refute, or bound a life-signature claim. We sell it as the second opinion the field has lacked: to observatories, research groups, and journals that need a verdict they can defend.

Visit site ↗
Since 202660% fellows · 40% commons
Computational Biology
Under researchIncubating

स्पंद (Spanda)

Creation and application of the digital cell. स्पंद (Spanda — Sanskrit for the primal “pulse” that animates living systems) builds computational, data-grounded models of living cells: virtual cells that can be perturbed, simulated, and queried, compressing wet-lab cycles for therapeutics, diagnostics, and basic biology.

Visit site ↗
Since 202660% fellows · 40% commons
Quantum Computation
Under researchIncubating

सूत्र (Sūtra)

Quantum error correction by discovery. सूत्र (Sutra — Sanskrit for the “thread” that binds, and the terse coded formula) searches the space of quantum error-correcting codes for ones that keep quantum information from decaying. A reproducible search found heavy-hex-native bivariate-bicycle codes that beat IBM's Gross [[144,12,12]] code on the field's figure of merit (k·d²/n) by 1.8–4.4× — one of them with its distance certified by an exact solver — and the same engine takes aim at the other open problems of fault tolerance: decoders, certified distance, and hardware-native code embeddings.

Visit site ↗
Since 202660% fellows · 40% commons
§ How a venture is born

Research becomes a venture only when someone outside the lab needs it.

  1. 01
    Incubation

    A line of research matures inside the lab. The compact already applies; no new paperwork is needed.

  2. 02
    Threshold

    An external party needs the work. A fellow proposes a venture. The cohort reviews and signs off.

  3. 03
    Launch

    The venture incorporates. The commons holds 40%; the founding fellows hold 60%, divided as they choose.

  4. 04
    Return

    As the venture earns, the commons' share funds the next generation of fellows. The loop closes.

Interested in licensing a venture, or bringing a problem that could become one?