Seeking the thermodynamic limit of biomanufacturing

Enter ↵ to ask Kairos · Shift+Enter for newline

✓ Opening Kairos with your question…

Built on AlphaFold DBStructure prediction Boltz-2Folding confidence screen PubMedLiterature retrieval & citation NCBISequence data source

Method

From question to design — one dry-lab loop

Kairos runs the design–build–test–learn cycle as a dry-lab loop today. Each stage declares what it is — retrieved evidence, model reasoning, or a deterministic tool.

01AskNatural-language research question› Natural-language question — e.g. express cytochrome c in Pichia, secreted?Live
02RetrieveCited evidence, with provenance› 66,000+ full-text papers · provenance on every cardLive
03DesignDeterministic CDS & leader panels› Deterministic CDS & leader panels — correct by constructionLive
04RankHost-fidelity scores you can re-run› Host-fidelity ρ ≈ +0.35 · scores you can re-runLive
05ReviewEvery claim audited — PASS / WARN / FAIL› 4 claims · 12 sources · verdict PASS (real dossier)Live
06FermentWet-lab validation closes the loop› Wet-lab validation via partner networkIn partnership
ASK RETRIEVE DESIGN RANK REVIEW FERMENT DBTL PICHIA-SCOPED

Product

One question in — a dossier you can audit

A Kairos session ends as a workbench dossier: cited evidence, a ranked design panel, structure checks, and a claim-by-claim audit. The demo below replays the published cytochrome-c run — every number real.

kairos workbench · live dossier
› Design a secreted expression construct for cytochrome c in Pichia
Research plan
Cytochrome c is a high-value heme protein — soluble recombinant production enables functional studies without mitochondrial isolation.
Signal peptide compatibility
Holo-cytochrome c maturation
Folding & strain engineering
AOX1 cassette optimization
About 1–2 minutesGo ahead →
Secretory signal peptide compatibility🚫 unproven
Holo-cytochrome c maturation in secretion🚫 unproven
Folding & strain engineering for secretion🚫 unproven
AOX1-driven expression cassette optimization🚫 unproven
Evidence too thin to call — Kairos says so instead of guessing.
Chat summary
Findingsecretion leader + host-weighted CDS ready; 12 synonymous candidates, host_weighted wins.
What doesn't helpevidence is thin on signal-peptide compatibility; all four hypotheses stay unproven.
94 evidence · 12 candidates · verdict PASS · 2 WARN
Open in workspace →
Evidence Design Structure Manufacture Audit
94 evidence · 6 retrieval rounds · EuropePMC 31 + KG 50 kept
Synergistic optimisation of expression, folding & secretion
PMID:33494776 · AppA phytase production in P. pastoris
Heterologous protein production in methylotrophic yeast
PMID:38730212 · recent progress review
Recombinant cytokines in K. phaffii
PMID:42113077 · design, production & purification
12 synonymous candidates · seed 42
#1host_weightedwinnerCopy CDS
#2gc_balancedtop tierCopy CDS
#3gc_balancedtop tierCopy CDS
Folding confidence screen
AlphaFold DB · 83 aa · mean pLDDT 91.2 — high confidence
lowpLDDThigh
Secreted construct · fed-batch AOX1 methanol induction
Centrifugation / microfiltration0.90
UF-DF (ultrafiltration / diafiltration)0.90
Spray drying0.85
Total recovery (multiplicative)0.689
Directional COGS $5–15/kg vs benchmark $5–50/kg · within parity range
Directional dry-lab estimates — absolute titer & recovery stay wet-lab.
verdict: PASS
7 claims · 94 sources — 5 PASS · 2 WARN
84 domain sources retrieved & citedevidence · EuropePMC/KGPASS
Candidate panel matches design outputdesign · deterministicPASS
Conversation-context statementreasoning · no sourceWARN
Signal-leader claim (P01149 / SUC2)target mismatchWARN
Generated Files · sources: EuropePMC ×31 · KG ×50 · every one linked

Evidence

Numbers we publish — and re-run

Dry-lab metrics we publish and re-run. Full methodology on /benchmarks/.

ρ ≈ +0.35

Within-protein host fidelity vs real titers (Spearman · 28 PichiaCLM variants, 5 proteins)

Minutes

From research question to ranked design panel

100%

Claims cited — then audited PASS / WARN / FAIL

Full metrics and methodology → /benchmarks/

Capabilities

What Kairos does

  • Ranks synthesis safety & host resemblance for Pichia constructs
  • Optimizes expression-construct elements — learned effects matching known biology (pCS1, ERO1/SBH1)
  • Scores 5′ translation initiation & protein developability
  • Checks structures in silico — AlphaFold DB + Boltz-2 confidence
  • Cites every claim to literature you can open and check
  • Audits every claim — automated PASS / WARN / FAIL verdict, line by line
  • Scores downstream manufacturability — fermentation, DSP recovery & directional cost, every number cited

Pichia-optimized dry-lab design, end to end — from cited evidence to an audited, order-ready construct. Bench validation runs through our partner network. Full scope statement →

Latest research

Deterministic designs, cited evidence — recent work from the Evolrix AI team.

View all research
Kairos adds an automated “review” layer — every citation traceable 2026-07-09 · Case study

Questions

What teams ask

What is Evolrix AI?

An AI company building the tools that make synthetic biology computable — dramatically accelerating the path from research question to manufacturing-ready design. Designs that are provably correct, scores you can re-run, and answers grounded in literature. Built for teams that need provenance, not persuasion.

Who is it for?

Industrial R&D teams in synthetic biology, metabolic engineering, and biotech — from early-stage design to pre-manufacturing validation.

What can it do today?

Kairos v0.19.6 (Pichia-scoped): conversational research with cited evidence retrieval, LLM-backed reasoning, ranked CDS design panels with host-fidelity scores and expression-construct optimization (element effects matching known biology (pCS1, ERO1/SBH1)), secretion leader design, a downstream manufacturability panel (fermentation, DSP sequence, step-wise recovery, directional cost — every number cited), hypothesis-driven verification where every hypothesis receives a supported / contested / unverified / insufficient verdict, and an automated review layer that audits every claim with a PASS / WARN / FAIL verdict and also assesses whether each cited source actually supports the claim (same organism, same target, entailed or contradicted). Expanding to more hosts.

How does the chat work?

Ask a research question in natural language. Small talk gets an instant answer; quick questions get a fast cited reply; research questions go straight into the full pipeline. Kairos retrieves cited evidence, reasons with an LLM, runs deterministic design tools, and streams the result into a multi-panel workbench — evidence, design, structure, manufacturability, audit — with a 1–2 screen summary right in the chat and a deduplicated provenance manifest (PMID / DOI / patent / knowledge graph) under Generated Files. Each section carries its provenance, and the audit panel shows a claim-by-claim PASS / WARN / FAIL verdict.

What is Kairos scoped to?

Kairos ranks dry-lab design quality — synthesis safety, host resemblance, expression-construct optimization, 5′ translation initiation, and protein developability. Expression and titer are wet-lab questions, answered by wet-lab data.

Bring a target protein. Leave with a ranked design panel

Minutes, not weeks. Every construct scored, every claim cited.