Blaise AI Beta: Entity-Linked AI for Medicinal Chemistry
Blaise AI is live in beta.
Not “chat, but with a molecule pasted in.” A workspace where the AI’s reasoning is attached to the objects chemists actually argue about: structures, series, assay tables, and the edits that connect one design decision to the next.
The real problem: chemistry work isn’t text
A medicinal chemistry program is a graph:
- Molecules (and their analogues)
- Assays (often messy, multi-source, inconsistent)
- Series hypotheses (“this substitution fixes lipophilicity but kills permeability”)
- Decisions (“drop the series”, “make these 3”, “add a solubilising tail”, “try a bioisostere”)
Generic chat treats all of that as a blob of tokens. The output might sound plausible, but it’s not anchored to the project reality.
Blaise is built around a different constraint:
Every claim the AI makes must point to the molecule(s) and the data it is using.
That’s what “entity-linked” means.
Entity-linked AI: the unit of reasoning is a molecular object
In Blaise, a message isn’t just text. It’s a set of references:
- a specific compound or congeneric series
- a specific assay readout (with provenance)
- a specific transformation (e.g., an MMP-style change)
- a specific hypothesis tagged to the series
The payoff is obvious the moment you try to do real work:
- You can click from a sentence to the structure it refers to.
- You can audit why the model said what it said.
- You can reuse conclusions later because they’re not free-floating prose.
SAR & series exploration that behaves like a project meeting
Blaise’s early beta emphasis is series navigation and SAR interrogation:
- Matched-pair analysis to surface transformations that actually moved the needle in your data.
- Assay-table navigation without the usual “copy/paste + spreadsheet archaeology”.
- A copilot that stays inside the series context instead of hallucinating global medicinal chemistry folklore.
A small-molecule-centric UI (not a chat box with RDKit sprinkled on)
The UI treats molecules as first-class objects:
- structures are selectable entities
- series are navigable units
- properties are attached to objects and provenance
This is chemistry-native interaction.
Who this beta is for
If you:
- live inside series decisions,
- spend time reconciling assay values across sources,
- want AI that is auditable and grounded,
…then the beta is already useful.
Try it
The beta is open with a free trial for early sign-ups.
Principle: the goal isn’t pretty text. It’s faster, more correct decisions about molecules — with a traceable chain of evidence.