We have been building AstraIQ for about two years. The core goal has always been the same: replace the email handoff between the simulation step and the lab step in inorganic materials R&D. Today we are inviting a small number of materials R&D groups to try the early version of what we have built.
This post describes what AstraIQ does, what state the software is in right now, and what we are looking for in early access partners. It is not a product launch announcement. It is an invitation to a handful of R&D teams to work with us on a problem we both care about.
The Problem We Are Solving
If you are running an inorganic materials R&D program, the discovery cycle probably looks something like this. A computational modeler generates a candidate list using DFT calculations or structure database queries. That list goes to the synthesis team by email, spreadsheet, or shared folder. The synthesis team picks from the list based on their own judgment about what is tractable given current lab capacity and precursor availability. They run synthesis, characterize results, and report back. Then the cycle starts again.
This process works. It has produced real discoveries. But its throughput is bounded by the coordination overhead at each step, and the feedback from synthesis back to the computational model is informal and often delayed by weeks. The modeler does not always know which candidates were actually synthesized, or why others were skipped. The synthesis team does not always know what the model expected to find, so they may not be measuring the right properties to evaluate success.
The email handoff is a symptom of a deeper issue: the simulation workflow and the lab workflow are not connected at the data level. Each runs in its own toolchain. Connecting them requires a shared representation of candidates, a defined protocol for dispatching synthesis based on screening output, and a feedback channel that updates the model when experimental results come back.
What We Built
AstraIQ has three components. The first is a screening engine that accepts a list of inorganic candidate structures (as POSCAR or CIF files) and evaluates them using a machine-learned interatomic potential trained on DFT data for inorganic crystals. The output is a ranked list of candidates ordered by predicted thermodynamic stability, with target property scores where applicable. The screening pass runs in minutes for a few hundred candidates, which is fast enough that it can be used interactively during the candidate generation phase rather than only as a batch post-processing step.
The second component is the synthesis dispatch interface. When a candidate list is approved for synthesis, AstraIQ generates a structured synthesis protocol payload that includes precursor identities, target stoichiometry, and a starting temperature profile based on prior synthesis data for the phase family. This payload is designed to be consumed directly by automated synthesis hardware, or to be used as a starting-point protocol for manual synthesis runs. The goal is to reduce the time between "this candidate looks promising" and "this candidate is in the furnace" from weeks to days.
The third component is the result feedback path. When characterization data comes back, it is logged against the candidate and the synthesis protocol used, updating the model's understanding of what the screening predictions actually mean for this composition space and this synthesis route. Over time, this calibration data improves the quality of future screening passes.
Where We Are Right Now
The screening engine is functional for oxide and phosphate cathode material classes. We have validated it against held-out DFT data for common NMC, LFP, and layered oxide composition families and the error bounds are acceptable for screening purposes, meaning good enough to rank candidates meaningfully and bad enough that DFT validation before synthesis is still warranted for any candidate you are seriously considering. We are honest about this limitation and it is built into the workflow design.
The synthesis dispatch interface currently produces structured protocol outputs that have been tested against two automated solid-state synthesis platforms. Support for additional hardware is something we plan to expand based on what early access partners are using.
The result feedback path is the least mature component. The data logging and candidate tracking work. The model update step that uses logged experimental data to recalibrate screening predictions is in early development. Early access partners will be running on a workflow where the calibration update is partially manual rather than fully automated.
Who We Are Looking For
We are looking for R&D groups at companies working on inorganic materials, specifically battery cathodes, oxide ceramics, or specialty phosphate compounds, who are running some form of computational screening today and who have at least partial synthesis automation or a high-throughput synthesis capability. We are not the right fit for groups that are doing entirely manual synthesis at very small scale.
We are also looking for groups who are willing to be candid with us about what does not work. The early access program is a technical partnership, not a product trial. We expect to encounter failure modes in the software, limitations in the synthesis protocol generation for specific material classes, and edge cases in the screening model that the current training data does not handle well. Our goal is to learn from those failure modes quickly and fix them.
Early access is not paid. We are offering time, the use of the platform, and direct access to the team in exchange for structured feedback on what the software does and does not do well for your specific use case.
How to Reach Us
If this sounds like a fit for where your R&D program is, reach out through the contact page and describe your material class and current workflow. We will respond to everyone who reaches out and have a direct conversation about whether early access makes sense for your situation. We are not looking for a large number of partners in this first phase. We would rather work closely with three or four groups than loosely with thirty.
The science page describes the technical approach behind the screening engine in more detail, and the platform page covers the workflow components. Both are works in progress, like the software itself.