eDNA, peptide and habitat signals captured at the source
Telemetry from live animals with no capture and no tissue loss
Models read the reads — variant calls, peptides and matches in minutes
Every specimen and sequence anchored to an immutable public record
Shared datasets, shared instruments, shared credit across labs
Decode nature's smallest signals. Genomic research, AI discovery, and transparent on-chain provenance — built for biology you can trust.
GeForce 8 · 2006
When NVIDIA launched the GeForce 8, the official technology demo they shipped was called Froggy — a photoreal frog you could poke, built to show what unified shaders could do. It still sits in their tech-demo archive. Twenty years on, the frog is no longer the benchmark. It's the subject — and the whole pipeline runs on the descendants of that silicon.
Amphibians are the most threatened class of vertebrates on Earth — roughly 41% of assessed species. They are also the source of some of biology's most useful machinery: the antifreeze strategy behind cold organ preservation, the skin chemistry behind a generation of antimicrobial peptides, and the only vertebrates that rebuild a limb from scratch. Each programme below targets one of those, and each one is limited by how much of the existing data anyone can actually process.
Programme 01
Continuous rainforest audio is being recorded far faster than anyone can listen to it — the Australian Acoustic Observatory alone holds around two petabytes under an open licence. We run species-detection inference across archives like it on NVIDIA GPUs — roughly a petabyte a week — turning an impossible listening problem into a tractable verification one.
Programme 02
Wood frogs in interior Alaska have been recorded frozen for 193 consecutive days, at hibernacula minima near −15 °C, with every animal surviving. The mechanism — solutes concentrated into a shrinking unfrozen fraction while ice is pushed outside the cells — is the model behind sub-zero organ preservation.
Programme 03
Magainins were found in 1987 because a surgeon noticed frogs returning to filthy water without infection. Since then amphibian skin has produced antimicrobials that reached late-stage human trials. The open question is selectivity: does a peptide destroy a bacterial membrane and spare a human one?
Programme 04
Chytrid fungus is the most destructive wildlife pathogen ever recorded, and the raw sequencing reads for hundreds of isolates sit in public archives with only a handful assembled into genomes. We assemble and annotate them against the current reference to resolve lineages and expose drug targets.
Programme 05
A twenty-four-hour drug exposure, delivered by a wearable bioreactor, has been enough to start an adult frog regrowing a functional hind limb over the following months. The signal is instructional rather than structural — which makes finding the right combination a search problem.
Programme 06
Assisted reproduction and cryopreserved sperm have already produced second-generation offspring for species down to their last few dozen wild adults. Every specimen we handle is registered, sequenced and anchored, so the record survives even when the population does not.
Field units collect environmental DNA, skin-peptide swabs, call audio and habitat telemetry without capturing or harming the animal.
Reads are basecalled at the collection point on each unit's onboard NVIDIA Jetson module and reported with their quality scores attached, so a result never travels separately from its confidence.
Alignment runs CUDA-accelerated on an NVIDIA H100 pod with Parabricks — a whole genome in under an hour — and peptide candidates are scored for selectivity before anything reaches a bench.
Specimen, sequence, instrument state and timestamp are hashed into a block. The finding and the conditions that produced it stay bound together.
41 SITES REPORTING
Reproducibility fails most often not because a finding was wrong, but because the conditions that produced it were never recorded precisely enough to repeat. Anchoring binds the two together permanently.
pH, temperature, dissolved oxygen and flow at the moment of measurement are captured with the reading, not written down afterwards.
A block commits to the specimen identifier, the sequence produced, the reference it was matched against and the quality achieved.
Once anchored, the record cannot be edited — only superseded by a later block that openly references it.
Recompute the hash from the published inputs. If it matches, the result is exactly the one that was recorded at the time.
Stated plainly, including the ones that are still simulation only.
| Programme | Target | Stage | Status |
|---|---|---|---|
| Bioacoustic surveyPRG-01 | Species detection over open archives | Production inference | Live |
| Specimen registryPRG-06 | Registered, sequenced, anchored specimens | Production | Live |
| Provenance ledgerPRT-00 | Immutable result anchoring | Production | Live |
| Peptide selectivityPRG-03 | Membrane-selectivity screening at scale | Validation against wet-lab panel | In progress |
| Pathogen genomicsPRG-04 | Lineage-resolved pangenome | Assembly backlog | In progress |
| CryobiologyPRG-02 | Ice-nucleation and ice-binding proteins | Simulation only | Research |
| RegenerationPRG-05 | Combinatorial cocktail search | Simulation only | Research |
The console is live: pick a specimen, drive the bioreactor, start a sequencing run and watch the result get anchored. Nothing is a screenshot.