The data exists. Finding and comparing it is the bottleneck.
Lab data lives in a LIMS, one or more ELNs, instrument exports and spreadsheets. Answering "have we run this formulation under these conditions before?" means searching notebooks and asking colleagues. When a scientist leaves, their context leaves too.
- Past experiments are hard to find across notebooks and systems
- Results summaries are written by hand from exported data
- Instrument data and sample records need manual matching
What we build
Experiment search
Searches ELN entries and LIMS records in plain language and returns matching experiments and samples, linked to the originals.
Results summaries
Drafts summaries across samples, batches or studies from structured data, for a scientist to check and edit.
Report drafting
Assembles recurring lab reports from LIMS data in your format, every figure traceable to its source record.
Data matching checks
Flags records where sample IDs, instrument exports and LIMS entries don't line up.
How it works
Step 1: Pick the questions that matter
We start with the searches and reports your scientists run most and where time is lost.
Step 2: Connect read-only
Scoped, read-only access to your LIMS, ELN and data stores. The tool never creates or modifies lab records.
Step 3: Show the source every time
Every result links to the underlying records, so a scientist can verify it.
Step 4: Keep sensitive data inside
For proprietary or regulated data, we can run models in your own environment so nothing reaches a hosted service.
Systems we connect
We haven't shipped LIMS or ELN connectors yet. We scope each system during Discovery.
- LIMS platforms
- ELN platforms
- Instrument data exports and shared data stores
- SharePoint, Microsoft 365 and Google Workspace
Reference
Our VP of AI Engineering, Michael Fethe, spent 15 years inside publicly traded agriculture and life science corporations. We bring that experience together with the documentation, audit-trail and private-deployment patterns we've built for regulated manufacturing and finance clients.
Related work outside life science: our reference architecture is an MCP server that lets AI query a contractor's construction ERP in plain language within defined permissions. The same read-only pattern applies to lab systems.
Questions lab leaders ask
Our data is proprietary. Does it go to an outside model?
It doesn't have to. Self-hosted and open-weight models can run in your environment. For hosted models, we use enterprise plans. Enterprise plans from Anthropic don't train models on your business data under their commercial terms.
Can summaries be used in GxP reports?
Only through the same human review and QA controls as anything else. Your QA team decides how the tool is validated for that use.