← Research

Research

The loop: read, reason, design, verify

June 28, 2026

A single inspectable loop from molecule to manufacturing. Every step carries its provenance — query, tools, arguments, results, checks, and sources kept in view.

Biology research does not fail for lack of intelligence. It fails because the path from a hypothesis to a verified design is scattered across tools, papers, spreadsheets, and a researcher's memory — with no persistent record of why each decision was made. Evolrix AI's research loop is built to keep that record.

It is a single loop with four steps. The output of each step is the input to the next, and every step emits an inspectable artifact — not just an answer.

  1. Read

    Retrieves across 66,000+ full-text papers and a curated knowledge graph — every claim traceable to its source. (Literature retrieval module: upcoming.) The reading step is not a summary; it is a set of grounded facts you can follow back to the literature.

  2. Reason

    Works step by step through verifiable tools you can inspect and challenge — not a black box. Each reasoning step records which tool was called, with which arguments, and what it returned. You see the chain, not just the conclusion.

  3. Design

    Turns a hypothesis into a concrete, validity-guaranteed design. It calls tools to produce sequences — it never invents them. A design is correct by construction: the candidate is verified to encode the exact target, in-frame, with no premature stop.

  4. Verify

    Scores feasibility in silico on literature-grounded, multi-objective criteria — before the bench. Verification is a deterministic computation re-runnable over the printed sequence, with per-axis explanations and surfaced risk flags.

An instrument log, not a press release

A research run should read like an instrument log: query, tools, arguments, results, checks, and sources kept in view. The interface is built around provenance, not persuasion. When Kairos recommends a candidate, you can see exactly which tools produced it, which axes favored it, and which flags it carries — including flags that argue against it.

Carried across every scale

The same loop applies at the sequence level and at the design-decision level. Tool-orchestrated design, literature retrieval, knowledge-graph reasoning, in-silico verification, transparent multi-step reasoning, and grounded design choices — each is a capability that keeps the loop intact as the problem scales.

Scope. The loop produces dry-lab design rankings and verified constructs, not expression or titer predictions. The verify step scores design quality — synthesis safety, host resemblance, 5′ initiation, protein liabilities. Real experimental outcomes await wet-lab data.