The reasoning layer above your materials knowledge
Choosing the right material is slow, and too often it comes down to luck: decisions made one property at a time, inside one industry, against a single supplier's spec sheet. Chemia changes that. Ask your real problem in plain language and Elixir reasons across the materials landscape to tell you what to evaluate, where each option is likely to fail, and what to test next.
+600,000 materials. Every value carries its source and our confidence in it.
The question that wouldn't go away
I came to physics for superconductors, and crossed an ocean to study them. What held me was a hard limit: after more than forty years, the temperature at which these materials lose all resistance has barely moved, and room-temperature superconductivity is still out of reach. The work is brutally hard, and most of it circles a few questions on a few materials.
During my master's I grew these materials by hand. It was slow, and worse, it was rarely clear which one to make next, so discovery too often came down to luck. The deeper problem was structural: in a funded lab you can only chase what you are funded to chase, so the field misses most of the possibilities. The bottleneck was not a shortage of brilliant people. It was that materials knowledge is scattered, and most of it is never connected. Life is short, and I did not want to spend it circling one question.
We started at the bench, building instruments to measure materials faster. That hardware lives on today through a university collaboration, but it is part of our history, not our roadmap. The real leverage was never another instrument. It was connecting what the world already knows. So Chemia became, first and foremost, an AI and data company.
The answer is often hiding in another field
Engineers do not ask what the thermal conductivity of a material is. They ask whether it will survive in their system, where it will fail, and what else they could use instead. Answering that means looking across fields, not down a single column of a datasheet.
The proof is everywhere once you look. To grow the single-crystal blades inside a jet engine, aerospace engineers learned to bend part of the growth chamber, a small trick known for years. The condensed-matter physicists growing their own crystals had never heard of it. In gas capture, humidity is the enemy, so teams design carefully around it; in semiconductors the same humidity sensitivity exists, but everyone is looking at electronic properties instead. The knowledge that solves one industry's problem is usually sitting, unread, in another's literature.
Chemia exists to connect that knowledge, and to move you from a value, a lookup, to what that value means for your decision.
Predict. Reason. Discover.
Elixir is not a database you search, and it is not a generator inventing molecules in a vacuum. It is a reasoning engine that works across industries, carries the provenance and confidence of every value, and shows its work.
Predict
Machine learning models fill the gaps across more than 600,000 materials, predicting physical properties even where no measurement exists yet. This is where our models are strongest today.
Reason
An ontology connects materials, properties, operating conditions and failure modes, so Elixir can weigh trade-offs and trace a failure back to its cause. Insulating power, for example, decomposes into porosity and bonding.
Discover
Constrained generative models propose new crystal structures, reaction candidates and composite designs as hypotheses. We are still testing this toward our reliability targets, and every candidate is checked with first-principles calculations before we trust it.
How we know what we know
A prediction is only useful if you can trust it, so every value in Elixir carries its origin and our confidence in it. We keep a clear order of evidence and never quietly mix a predicted number with a measured one:
- Measured
- ›
- Computed
- ›
- Model-predicted
- ›
- Literature-extracted
We validate on held-out data. For the dielectric constant, for example, we trained on first-principles data and tested on a set held out by symmetry group, then checked the model a year later against newly available materials it had never seen. We see the same behaviour on other properties such as specific heat. We are glad to walk a technical team through the details.
We are also clear about what we do not do. Elixir does not certify materials, and it does not replace your testing or simulation. It changes what you choose to test. Predictions come with confidence bounds and a recommended validation; generated structures are hypotheses, not discoveries. Our scientists stay in the loop where it matters: validating the data our pipelines extract from commercially compliant sources, and checking predictions against symmetry groups, known physics and sanity tests so the models keep improving.
And we are careful with knowledge. What is public, across every industry, feeds a shared knowledge base that stays available to everyone. What a client shares privately stays private, used only to sharpen their own edge and never folded into the common pool. We do not hoard public knowledge, and we do not leak private knowledge.
We have already put this to work. In a clean-energy pilot, a partner needed materials with the right selectivity at a specific temperature and pressure for energy storage. Data was scarce, so beyond our predictions we ran grand canonical Monte Carlo and first-principles calculations. For several harder classes, such as metal-organic frameworks, our results landed within the measured range on the samples we checked, including sanity checks against the partner's own proprietary data, and we extended the work into predictions and property classification for a much wider set of materials.
Pilot projects and discussions are now in progress with companies across these industries and in academia.
Chemia has grown with the support of programs like Next AI, Plug & Play and ACET.
Leadership
Team
Former Interns and Collaborators
- Étienne Lacroix
- Mathieu Labbé
- Frédérick Messier
In the Community
- Frédérick Messier
- Étienne Lacroix
- Gabriel Gaouette
- Simon Lefebvre
- Coralie Pelletier-Ouellet
- Williams Gravel
- Amaury Daniel Palao Garcia
- Mathieu Labbé
We want to take the guesswork out of materials
If we get this right, engineering design stops running on guesswork. Elixir tells you early, and with reasons, which materials and designs to avoid because they will fail, and which will get you where you need to go. Then it goes further and points to what is newly possible: a material that lasts longer, costs less, or unlocks a product that could not be built before.
We help the teams who refuse to settle for a generic material, in semiconductors, clean energy, aerospace and any field where pushing material performance and reliability is the whole game. Beyond saving them time, we want to keep raising the baseline of what materials can do, and help move whole industries toward their next breakthrough.