Skip to main content
Labvana
Scientific AI InfrastructureMaterials & chemistry R&D

Building AI that accelerates scientific discovery.

Turn fragmented R&D data into reproducible, AI-driven workflows — cutting the time between an experiment and the decision it informs.

Watch the DemoPartner With Us

Founded by researchers from Berkeley Lab and STMicroelectronics.

From problem to solution

Research data is scattered. We connect it to the decision.

Today the call comes down to expert judgment alone, because the evidence sits across instruments, notebooks, and spreadsheets that never meet. VanaNexus brings it together — so decisions are driven by what you measured as well as by what you know.

Scattered data

Instruments, spreadsheets, and notebooks that never meet.

Unreproducible workflows

Results nobody can rerun six months later.

Lost context

Knowledge that leaves with the person who made it.

EVERYDAY LAB KNOWLEDGE & DATATRUSTED INSIGHTS & FASTER DISCOVERYResearch Papers & PatentsElectronic & PaperLab NotebooksSpreadsheets &Experimental DataRaw Instrument Outputs& MeasurementsExisting Code &Simulation ResultsVanaNexusYour Connected ScientificIntelligence HubReproducible & ReplayableExperimentsEnd-to-End TraceabilityFrom Raw Instrument Datato Final ResultsClean, Connected DataReady for Analysis & AIScientific AI Agents Groundedin Verifiable EvidenceFaster Experiment Cycles &Fewer Wasted Trials

How it works

Every decision stays connected to the run that produced it.

One traceable line from raw instrument data to the recommendation a reviewer can audit.

  1. 01

    Data

    Captured with schema and lineage

  2. 02

    Workflow

    Versioned, rerunnable steps

  3. 03

    Model

    Trained on real experimental context

  4. 04

    Decision

    Traceable back to the raw run

Inside the platform

This is the product, not a mockup.

Screens from VanaNexus, the Labvana platform in active development.

The VanaNexus workflow canvas showing a 16-node Concrete Strength Random Forest Training and Evaluation workflow — load_csv, drop_nulls, and standardize branching into two train/test splits that feed random_forest and gradient_boost_tree models, then ml_metrics, plot, and csv_write — with the project list on the left and run #626 recorded as successful on the right

Build it once, rerun it forever

Sixteen nodes: load, clean, and standardize the data, split it two ways, train a random forest against a gradient-boosted tree, then score, plot, and export both. Run #626 is on the right, recorded and rerunnable.

Running today across

Cheminformatics and molecular MLPredictive modelling and evaluationBayesian optimization and experiment designQuality, measurement, and statistics

Product demo

See it run

Three recorded walkthroughs: the platform overview, building a workflow, and handing work to an AI agent.

  • Workflows you can rerun, not chat you have to redo
  • Every step carries its data and model context
  • Built for teams without an in-house ML group

What VanaNexus is and who it is for.

Let's build better scientific infrastructure.

We work with researchers, engineers, and organizations building the next generation of scientific tools.

Research institutionsIndustrial R&D teamsTechnology partnersDesign partnersInvestorsOpen-source contributors
Contact Labvana