AI Scientist

A scientist. Not a chatbot with a lab coat.

Eidosoma AI Scientists are always-on research agents that scale the work of the humans they work for. They read the literature for you, propose hypotheses, write and run hundreds of computational experiments in parallel, and surface the few that matter — letting a small team operate at the scale of a large one, without losing the agency, taste, or judgement of the scientists they serve. We set these up for partner labs as part of our consulting engagements — every AI Scientist is built, tuned, and handed over through the same process.

Principles

Five non-negotiables.

  • 01Scales your output — one scientist running experiments at the pace of dozens, and growing
  • 02Entirely owned: every line of code is your intellectual property
  • 03Designed to empower — not replace — human scientists; your ideas, your taste, your direction
  • 04Dynamic papers that issue updates and re-open for new experiments
  • 05Runs continuously; hands you mornings-worth of findings each day
Modules

Built from composable research modules.

Each Eidosoma AI Scientist is assembled from a set of composable modules — below are three of the core ones. Tap to unfold.

01
Collective Intelligence Processor
Daily large-scale web & literature sweep for every active project.

The CIP reads the internet the way a lab reads coffee: every morning, in large quantities. It crawls preprints, patents, GitHub commits, conference schedules, patient registries, and lab-notebook blogs — then filters, cross-references, and rewrites findings as context for your ongoing projects. Novelty, surprise, and contradiction are prioritised over volume.

02
Experiment Coding Manager
Runs a team of AI coders that build and execute computational experiments.

The ECM decomposes a research question into runnable experiments, spawns a team of specialised AI coders, reviews their pull requests, and orchestrates execution on the cluster. It keeps a full experiment graph — every run, every parameter, every negative result — so that nothing is ever re-discovered by accident.

03
Evolution Engine
Optimisation through genetic algorithms and open-endedness — MAP-Elites, NSLC, quality-diversity.

Good science rarely comes from climbing the steepest gradient. The Evolution Engine maintains a living archive of diverse candidates — models, protocols, organisms, hypotheses — and keeps exploring the space of what has not yet been tried. Underneath: MAP-Elites, novelty search, and open-ended divergence.

A day in the life

What your scientist does while you sleep.

22:00
Runs overnight literature sweep. 723 papers scanned, 24 flagged as relevant.
01:30
Generates four new hypotheses from flagged papers. Three survive internal review.
03:00
Queues 46 computational experiments across three projects. Begins execution.
06:45
Drafts two patch notes for your dynamic paper. Flags one experiment as suspicious.
07:30
Writes your morning brief. Three findings, two questions for you, one request for a wet-lab replicate.
08:00
You open your laptop.

Work with us on a living science.

We are planning initially to partner with a small number of labs and companies this year. Consulting engagements begin with a two-week discovery sprint.