Decision-first AI
for battery materials discovery
ASTR-1 accelerates battery materials discovery by helping researchers decide which candidates to advance, repair, reject, validate, or prepare for synthesis.
A deep-tech company at the intersection of AI, materials science, and quantum-inspired computing.
Neognos Tech exists to make battery materials discovery more systematic, explainable, and decision-driven. We are building the architecture that turns enormous candidate spaces into a sequence of well-justified decisions — advance, repair, reject, validate, or prepare for synthesis — so research time is focused on the candidates that deserve it.
From generating candidates
to deciding on them.
Generative models can produce millions of plausible battery materials. The harder question — the one that decides whether research time is well spent — is which of those candidates deserve a researcher's next hour, next experiment, next gram of precursor.
ASTR-1 is built around that question. It treats discovery as a sequence of governed decisions, each one supported by the right kind of evidence at the right cost.
Battery materials discovery is a decision problem in disguise.
Modern AI can generate candidate materials almost without limit. The bottleneck has moved upstream — to the question of which candidates deserve a researcher's time.
Vast candidate space
Battery-material research spans an enormous combinatorial space of compositions, structures, dopants, and processing routes.
Evaluation is costly
Each candidate consumes compute, reagents, and skilled researcher attention. Naive parallelism is not a strategy.
Many failure modes
Candidates fail for instability, poor performance, toxicity, element scarcity, cost, or sheer synthesis difficulty.
Deciding is the bottleneck
The hard problem is not generating candidates — it is deciding which candidates justify further effort.
Traditional research pipelines treat generation, prediction, and validation as loosely coupled steps. ASTR-1 treats them as a single controlled evidence workflow, where every handoff is gated on evidence appropriate to the risk.
One architecture,
tuned for energy storage.
ASTR-1 is not a general-purpose materials generator. It is shaped around the specific failure modes, validation costs, and synthesis constraints that govern modern battery chemistries.
- Battery-first scope — cathodes, anodes, electrolytes, interfaces.
- Decision-first orchestration via the Master Route Controller.
- Uncertainty-aware gating between every stage.
- Failure-learning flywheel built into the pipeline.
Built for decisions, not just candidates.
ASTR-1 differs from generic generative pipelines in how it governs the journey from candidate to commitment.
Decision-first architecture
ASTR-1 is organised around the decision a researcher needs to make — not around a single model or a single dataset.
Governed AI workflows
A Master Routing Controller orchestrates the seven internal pipelines and decides when to escalate, repair, or reject.
Evidence-based progression
Candidates move forward only when the evidence supports it. Low-confidence outputs are escalated, not silently accepted.
Traceable scientific decisions
Each candidate carries a record of which pipelines ran, what they returned, and which gate let it advance.
Human-guided validation
Researchers stay in the loop. Physical testing is permission-gated and the workflow surfaces context, not just answers.
Failure-learning flywheel
Negative results are first-class. Every failure feeds back into routing so the next decision is better informed.
A network of specialised pipelines,
one routing brain.
Seven internal pipelines each produce a distinct kind of evidence. The Master Route Controller weighs that evidence and decides what happens to a candidate next: advance, repair, escalate to higher-fidelity validation, or reject and learn.
The result is acceleration that stays accountable — every decision carries a trail of what was checked and why.
An interactive view of the ASTR-1 decision pipeline.
From an initial research request to a reported decision — each node produces evidence the controller uses to decide what happens next. Hover or tap a stage.
- 01Request
- MRCMaster Route Controller
- 02Generate
- 03Uncertainty Gate
- 04QCVS (Validation)
- 05Dopant Prediction
- 06Property Prediction
- 07Result Capturing
- 08Reporting
- 09Optional Robotic Synthesis
A research goal enters the system — target chemistry, application envelope, and constraints define the run.
Five intersecting disciplines.
ASTR-1 is the artefact, but the research direction is broader.
AI
Learned models for property prediction, routing, and repair across the candidate journey.
Battery Materials
Cathodes, anodes, electrolytes, and interfaces — the chemistries that define modern energy storage.
Quantum-Inspired Computing
QIGA explores candidate spaces using quantum-inspired formulations, implemented classically.
Materials Informatics
Structured representations and failure databases that turn negative results into reusable knowledge.
Sustainable Energy
Prioritising earth-abundant, recyclable, and lower-impact chemistries in scoring and feasibility gates.
Vision, mission, ambition.
What we are working toward, and the discipline we are working with.
A discovery layer for clean energy.
We imagine a future where battery-material discovery is not a decade-long obstacle to clean energy adoption but a predictable, evidence-driven research practice.
Make every decision count.
Build the controlled evidence workflow that lets small teams of researchers responsibly explore candidate spaces they could never explore by hand.
Honest acceleration.
No exaggerated claims and no quantum hardware. Quantum-inspired methods, hybrid prediction, and uncertainty-aware validation — applied where they genuinely help.
The stack behind ASTR-1.
Eight building blocks composed into a single decision-first architecture.
Quantum-inspired generative algorithms
QIGA candidate generation across composition and structure spaces, classical implementation.
Hybrid property prediction
Learned models combined with physics-informed estimators at multiple fidelity levels.
Feasibility screening
Stability, availability, toxicity, cost, and known-failure pattern filters.
Uncertainty-aware routing
Calibrated uncertainty drives whether a candidate is accepted, escalated, or repaired.
Dopant and composition repair
Targeted modifications to fix specific weaknesses rather than discarding promising candidates.
Synthesis planning
Candidate synthesis routes and precursors aligned with practical lab constraints.
Failure database
Negative results are captured and reused — the learning flywheel of ASTR-1.
Permission-gated physical testing
Optional handoff to physical or robotic labs, opt-in and operator-governed.
Notes from the research direction.
Short writeups on the ideas behind ASTR-1. Full explanations live on the Research Direction page.
"Acceleration without accountability is just noise at scale. ASTR-1 exists to make every decision in a discovery run traceable to evidence."
Neognos Tech · Research direction
Start discovering smarter materials.
Explore the ASTR-1 architecture, or talk to us about research collaborations and evaluation partnerships.