Protova · Design

Search every protocol, and show the working

Protova agents propose candidate designs, run each one through Insilon, and return a ranked list with the trade-offs written down for your team to review.

Example design searchExample

Dose and sample size, phase 2b

Candidates scored
312
Top design
40 mg, n = 180 per arm
Predicted power
86% (82–89%)
Equity check
Age 75+ within 5% of overall
Reviewer
Required before export
3 designs shortlisted for reviewDraft
Example. Values are illustrative.

Capabilities

What Protova does

Search

Multi-agent design search

Agents explore dose, cohort criteria, endpoints and sample size within limits you set.

Scoring

Simulation in the loop

Every candidate is simulated, then scored on power, efficacy, safety and subgroup equity.

Ranking

Trade-offs written down

Each ranked design says why it beats the next one and what it gives up.

Drafting

Evidence text with sources

Drafts the MIDD rationale with citations to runs and outcome data.

Guardrails

Bounded agents

NeMo Guardrails limit agents to validated simulations and approved sources.

Review

People decide

Nothing leaves the platform without a named reviewer's approval.

How it works

Inside Protova

  1. Frame

    Set the decision, limits and objectives.

  2. Propose

    Agents generate candidate designs.

  3. Score

    Insilon and Vigila evaluate each one.

  4. Rank

    Return a shortlist with reasons.

Specification

Specification

Items marked planned are in design with pilot partners.

ItemDetail
Search spaceDose, schedule, inclusion criteria, endpoints, sample size, randomisation ratio
ObjectivesPower, efficacy, safety, subgroup equity, cost proxy
AgentsNeMo agents with Guardrails; tool use limited to Sanenza runs
OutputsRanked designs, rationale drafts, run references
AutonomyProposes only; approval required
StatusIn development

Questions

About Protova

Does Protova write the protocol?

It drafts the design rationale and evidence text. Your team writes and owns the protocol.

Can we constrain the search?

Yes. Set hard limits on dose, sample size, endpoints and inclusion criteria before the search starts.

How do we know the agent did not invent a source?

Every claim in a draft links to a run or an outcome record. Text without a link is flagged for the reviewer.

See Protova in a briefing

We walk your team through Protova on a public example, then discuss where a pilot could test it against outcomes you already know.