VALTRIX
Audit-Aware AI for Scientific Screening and Materials Discovery
VALTRIX helps scientific R&D teams prioritize which materials and molecular candidates deserve costly validation through DFT, laboratory testing, or deeper experimental review.
Benchmark-validated on Matbench. Submission-ready on JARVIS. Built for reproducible candidate prioritization.
The bottleneck is not only candidate generation.
Scientific discovery increasingly generates large candidate spaces. The expensive decision is deciding what deserves validation. Wrong prioritization wastes compute, lab time, and R&D budget.
Prediction plus audit discipline
VALTRIX is designed for reproducible screening workflows: data governance, benchmark discipline, and claim-safe evidence reporting.
What VALTRIX does
VALTRIX is an audit-aware AI screening platform built to support candidate prioritization, scientific ML evaluation, and reproducible decision workflows.
Prioritize candidates
Rank materials or molecular candidates before expensive validation, with the goal of improving triage quality under fixed R&D budgets.
Keep the trail intact
Support evidence packages, fixed splits, reproducibility controls, and audit-oriented reporting rather than isolated model scores.
Make claims defensible
Separate what is proven, what is supported, and what remains pending, reducing overclaim risk in scientific and partner-facing work.
Evidence Snapshot
The current evidence stack is presented with explicit claim boundaries. VALTRIX is not claimed as a universal discovery engine or a public leaderboard winner across all benchmarks.
External benchmark evidence
V7Core++ outperformed the project’s descriptor baseline under the official Matbench protocol.
V7Core++: 32.90 ± 9.65 MAE
Descriptor baseline: 38.71 ± 10.59 MAE
Submission-ready pathway
A regression-safe submission pathway has been prepared for exfoliation-energy prediction. External leaderboard outcome remains pending.
Industrial-style evidence
MOF work supports the broader screening narrative through blind/OOD-style evidence, robustness checks, and candidate prioritization framing.
Use cases
Layered-material screening
Candidate prioritization for exfoliation-energy and related property prediction workflows.
ADMET and chemistry workflows
Governed scientific ML workflows where split integrity, scaffold awareness, and reproducibility matter.
Decision infrastructure
Support teams that need a defensible ranking and evidence trail, not just a raw prediction score.
Technical and strategic partners
- Materials or chemistry R&D groups
- Research collaborators
- Grant and non-dilutive funding routes
- Deeptech investor conversations