Material science, but with a team. Eight specialists, one research plan.
A small Research Director ANN plans the workflow. Then Literature, Simulation, Physics, Design, Optimization, Experiment, and Validation ANNs do their part. Each is published, versioned, and rateable in the Aigarth marketplace. This page is the build-in-public view of the proposal.
The team
Eight small, specialized ANNs collaborate to discover and validate a new material. Each one excellent at exactly one part of the research workflow.
Research Director
Reads the research question, plans the workflow. Returns a structured JSON describing stages, costs, time, and constraints.
Literature
Ingests scientific papers. Extracts composition, property values, measurement method, citation. Builds a knowledge graph.
Simulation
Runs the actual science. DFT relaxation (VASP, Q-E), molecular dynamics (LAMMPS, GROMACS), or fast ML surrogates (MACE).
Physics Reasoning
Sanity-checks predictions against first principles. Catches impossible structures and runaway formation energies.
Material Design
Generates candidate materials. New alloys, new polymers, new battery cathodes. Output is a structure + predicted property + uncertainty.
Optimization
Finds the best tradeoff. Multi-objective Pareto on (energy density, cost, cycle life, environmental impact, ...).
Experiment Planning
Converts a candidate material into a lab protocol. Synthesize by solid-state reaction at 950 °C for 12 h. XRD, SEM, cycling.
Validation
Compares predictions against reality. Cross-checks against literature. Logs the result so the next iteration starts from a better baseline.
How a discovery flows
A research question becomes a plan. Each stage consumes the plan and produces an artifact. The Physics Reasoning ANN in the middle decides whether to flag a stage for re-run.
Every stage emits a structured prediction with uncertainty. A bad stage can be re-run alone. The rest of the workflow is preserved.
The 8 ANNs are live today
Each of the 8 roles is a real, published, deployable ANN in the Aigarth marketplace. Stub backends today; the real Director, Literature, and Experiment Planner are LLM calls; the rest are wired to the existing compute fabric.
Will it fit the Aigarth architecture?
We evaluated four dimensions. Most of the platform is ready. The continuous-learning pillar is greenfield, and the most important one for honest scientific improvement.
Registry, versioning, reviews, marketplace listings, deploy hook: all ready. Add `material_domain` enum, knowledge-graph tables, and 9 new endpoints under `/v1/materials/*`.
services/compute is task-agnostic. A DFT relaxation is a job like a video render. But material science is 50–100× more expensive, needs GPU as a first-class resource, and requires capability-tagged workers.
No per-ANN quality signals today. Material science adds per-prediction provenance, validation against experiments, uncertainty calibration, and *negative* results as first-class signals.
Qubic's outsourced-computing contract is *especially* suited to material science. Same K12 signature verification blocker as the video proposal: solve once, reuse.
Roadmap
Five phases. Re-evaluate at the end of Phase 1 with 1,000 ingested papers and a research question answered in under 5 minutes.
Architecture evaluation, blog article, dashboard, tracker. No code in production.
Paper ingestion + knowledge graph + research-question answering. No simulation. Re-evaluation gate: 1,000 papers ingested, answer in < 5 min.
Wire up DFT, MD, and MLIP workers. End-to-end material discovery. ~52 QU per workflow, ~102 h wall-clock on 1× CPU.
Publish the 8 ANNs as real listings. Per-ANN quality ranking. Reviews and ratings flow into the feedback table.
On-chain ANN ownership, worker registry, reputation, rewards, and `discovery_attribution`. Off-chain rollups + one-way publish.
What it costs (illustrative)
Find-a-new-battery-cathode workflow on a single CPU worker. Numbers are placeholders that will change as we ship and measure.
| Stage | Cost (QU) | Wall time |
|---|---|---|
| Literature (30 papers) | 0.10 | 5 min |
| Research Director (plan) | 0.05 | 1 min |
| Material Design (100 candidates) | 1.00 | 30 min |
| MLIP screening (MACE) | 5.00 | 60 min |
| DFT refinement (VASP, top 10) | 45.00 | ~100 h |
| Physics Reasoning | 0.20 | 5 min |
| Optimization (Pareto) | 0.10 | 1 min |
| Experiment Plan | 0.05 | 2 min |
| Validation | 0.05 | 1 min |
| Total | ~51.55 | ~102 h |
Cost is dominated by DFT refinement (87%). MLIP surrogates cut the wall time by 10×. A monthly reservation of 1,000 QU covers ~20 cathode discoveries per month.
The single hard precondition
Everything in Phase 4 (the on-chain portion) depends on one thing we haven't shipped yet.
services/qubic is currently format-only. The address regex passes, but the cryptographic signature isn't actually checked. This blocks every on-chain write in the proposal, including the new discovery_attribution contract for material findings. The fix is to use @noble/curves (already a dep) for real K12 verification. ~3 story points of focused work, no new architecture. Same blocker as the video proposal: solve once, reuse.
Follow along
This is a build-in-public project. The dashboard at localhost:4000/material-science is the source of truth for what is real today.
Numbers, phases, and product surfaces in this article are illustrative placeholders that will change as we ship and measure. Aigarth Cloud is a build-in-public project; the dashboard is the source of truth.