Chain-of-custody forms in. Reviewed sample-login drafts out.
SampleIntake is an AI intake module for LIMS vendors. It turns scanned, photographed and handwritten COCs into source-linked sample-login drafts that your labs' technicians review and approve, then imports them through your existing import path. Nothing enters the LIMS without a person approving it.
Review screen: every field links to where it was read on the page; uncertain fields are flagged, never guessed.
How it works
Manual COC transcription is the most visible daily labor cost in environmental, water and food labs. SampleIntake handles the reading and checking; your lab's staff keep the decision.
Every field, with its source
Header, every sample row, analysis checkboxes or free text, custody hand-offs and the receipt block, from PDFs, scans or phone photos. Each value keeps a link to where it was read.
To the lab's own catalog
Requested analyses map to the lab's test codes; dates, matrices, preservatives and drinking-water fields (PWSID, sample type, chlorine residual) are normalized. Reviewer corrections teach it the lab's vocabulary.
SOP rules before login
Hold times exceeded or at risk, temperature on receipt, preservative vs method, unsigned custody, tests not in the catalog, missing chlorine residual, and missed sample rows.
Review, approve, import
Keyboard-first review of flagged fields only, then approval and export in the format your import path reads, with the original COC attached and a full audit trail.
Built to sell through your LIMS
Not another LIMS. An add-on your customers get from you, delivered through the integration you already support.
- Your customer relationship and your system of record stay yours.
- Exports to your import format: CSV, fixed-width, XML, JSON or SQL Server staging tables, configured from your import spec. No rewrite of your product.
- Configured per lab from your test-code export: catalog, SOP rules and export mapping.
- Hosted, or as a Docker image on your or the lab's servers.
Built for the people at the bench
The work is review, not retyping.
- Calibrated confidence: fields are auto-accepted only when they are very likely correct.
- Side-by-side page and draft; Tab jumps between flagged fields.
- A second independent read catches missed rows and disagreements.
- Every change logged: who, what, when, and from which value.
Evidence so far
Measured on real blank COC forms published by 25 US labs and state programs, filled with realistic handwriting and scanned or photographed for testing, scored on forms the system was never tuned on.
These are pre-pilot test results, not production numbers. A pilot measures what matters on a lab's own COCs: end-to-end intake time (assisted vs manual, randomized, same screen and timer) and critical-field errors.
Data handling
- Customer documents are confidential and isolated per lab.
- The original COC is stored unchanged; every edit is in an append-only audit log.
- Named user logins; the audit trail records the authenticated reviewer.
- We don't train models on lab documents.
- Self-hosted deployment available for labs that keep data on their own servers.
About
SampleIntake is built by specialists in large language models, computer vision and speech.
We work with LIMS vendors to bring modern document AI to the labs they already serve, without replacing the systems those labs rely on.
A two-week pilot with one of your labs
One lab, up to two COC form types, your supported import path. Measured on total intake time and critical-field errors on the lab's own held-out COCs. Before we build, we agree what continuing costs if it works.
admin@sampleintake.com