Biotechnology AI agent · Quality control
LIMS & Sample-Tracking Agent (Custody to Result)
Follow a sample from receipt to result — custody, location, storage excursions, aliquot parentage, timepoints and turnaround — and hold what will not confirm, for your analyst, QC supervisor and QA to act on.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Result release
- QC and QA only
- After launch
- Agent Care
What this agent does
Follows the sample, from login to result
In
- 01
Take the sample as it arrives — the request, the labels, the custody record and the conditions it travelled in.
- 02
Read the LIMS as the system of record — logins, locations, aliquots, storage, timepoints and test assignments.
Reason
- 03
Check the login against the request: identity, containers, condition on receipt and the tests asked for.
- 04
Follow custody, location, aliquot parentage and consumed volume, and mark where the chain does not join up.
- 05
Set storage readings, timepoints and turnaround against the protocol and the specification in force.
Decide
- 06
Route a reading outside its limits into the investigation procedure your SOP defines, to a named person.
- 07
Hold anything whose identity, custody or record will not confirm, and say what is missing.
Out
- 08
Present the sample's history, the gaps found and the record standing behind each result.
- 09
Put an excursion in front of a person with the chamber log and the product's own stability data beside it.
Product statement
The agent tracks, compares and routes inside the approval boundaries agreed during implementation, and writes carry a trail that is attributable and reviewed. Releasing a result, calling one invalid, authorising a retest or a resample and disposing of a sample stay with QC and QA.
Example workflow
One sample, end to end
- 1Sample received and loggedThe request, the container labels, the condition on receipt and the custody record that came with it
- 2Identity and custody checkedLogin set against the request: identity, container count, aliquot parentage and every move in the chain
- 3Conditions and timepoints readStorage readings against the protocol limits, stability and retention timepoints, expiry and consumed volume
- 4Tests and record assembledTest assignment against the specification in force, turnaround against the plan, and the record behind each result
No human action required
Stages 1 to 4 run without a person in the loop — nobody is asked to read a sample file until custody, conditions and the record behind the tests have been checked. A reading outside its limits leaves for a named person the moment it is seen.
Goes to the analyst as a checked sample file.
Held with the unconfirmed link marked.
Analyst, QC supervisor and QA
Read the history, the excursion and what would not confirm, then run the investigation your procedure defines — a result is invalidated on a cause that investigation finds, not on a retest that came back cleaner.
Accept · Correct · Investigate
Accepted — handed back
- 6LIMS updatedWritten only where access and policy allow, with an attributable trail; no result is approved and none released
- 7Outcome evaluatedExcursions missed, flags a supervisor dropped, records still short at review and corrections by cohort
Corrections
Flags a supervisor drops or rewrites are counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Approving or releasing a result.
- Deciding an out-of-specification result is invalid.
- Authorising a retest or a resample.
- Disposing of, consuming or reassigning a sample.
Automation boundary
Agent acts unaided
- ✓Check the login against the request and the labels.
- ✓Follow custody, location, aliquot parentage and consumed volume.
- ✓Compare storage readings, timepoints and turnaround with the protocol.
- ✓Flag the gaps and hold what will not confirm.
Write actions run only inside the approval boundaries agreed during implementation. Releasing a result is not among them.
- Changing a specification, method or stability protocol.
- Accepting a storage excursion as having no impact.
- Closing an investigation or naming its cause.
- Changing the tracking rules, limits or thresholds.
Example output
One stability pull, annotated
Everything the agent flags is attached to the sample and the record it came from.
Sample file · single stability timepoint
Illustrative example- Sample
- Stability pull, mid-point
- Storage
- Excursion during a transfer
- Custody
- Unbroken
- Proposed
- Assess against stability data
- Confidence
- 92%
- Result status
- Left for QC and QA
As received
The pull, its timepoint and the conditions it was held in, as the LIMS and the chamber log hold them.
Evidence used
- Chamber log for the period
- Custody and aliquot chain
- Stability protocol timepoint
Why it is flagged
The reading left its limits in transfer, so this product's own stability data decides.
What the score decides
Confidence decides how much a supervisor re-checks, not whether a result may stand.
Action
AcceptCorrectInvestigateValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every sample in the laboratory
From receipt to disposition
Keep identity and custody joined up
Use the request, the labels, the custody chain and the specification in force — identity is what every result above it rests on.
Stop rebuilding the sample's history
Custody, location, aliquots, storage and timepoints arrive assembled, so an analyst starts from the sample file rather than from five screens.
Raise what a result cannot rest on
Storage excursions, broken custody links, missed timepoints and records short of what a result needs reach a named supervisor as a list.
Build an evidence trail
Retain the readings taken, the rule behind each flag, the custody links checked, the confidence and every supervisor and QA edit — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
LIMS & laboratory systems
- LabWare
- LabVantage
- STARLIMS
- SampleManager
- laboratory APIs
Instruments & chromatography
- Empower
- Chromeleon
- OpenLab
- Balance and reader feeds
- audit logs
Storage & environment
- Chamber logs
- freezer monitoring
- Excursion alarms
- shipment loggers
Agent
LIMS & sample tracking
- Checks identity
- Follows custody
- Routes to a supervisor
Quality & manufacturing
- TrackWise
- Veeva QualityDocs
- SAP / MES links
- stability schedules
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the sample record
Each control wraps the one inside it. A check clears every layer before it reaches the sample file, and the trail it leaves is written to be reviewed, not only kept.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether a check may stand
- L3
- Routes a reading to the procedure, and is measured on it
- L4 – L5
- Leave release to QA, keep the trail reviewable
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the whole sample file — not only the flag at the end.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the sample file the analyst opens
- E1Final-output evaluationWas the flag real, and the rule it names the one in force?
- E2Custody evaluationDid identity, aliquot parentage and every move join up?
- E3Condition evaluationWere readings and timepoints read against the right protocol?
- E4Routing evaluationDid a reading outside its limits enter the procedure, and how fast?
- E5Slice evaluationHow does performance change across specific sample cohorts?
- E6Business outcomeHow much was re-checked, and what was still missing at review?
Floor — the result a supervisor and QA stand behind
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle. The most cited of them is a result treated as invalid because a repeat came back cleaner, rather than because an investigation found a cause.
Agent lifecycle
Direction of processing →
01 · Intake / login
2 modes
- LS-01
Custody gap reads as a clean chain
A move with no record reads as continuous, so identity stops being provable.
- LS-02
Aliquot inherits the wrong parent
A child tube is linked to the wrong parent and carries its history.
Stage gathers
The request, the labels, the custody record and the LIMS
02 · Tracking
1 mode
- LS-03
Excursion smoothed into a period mean
A spike is averaged away instead of assessed against the stability data.
Stage compares
Custody, conditions and timepoints against the plan
03 · Assessment
2 modes
- LS-04
Reading outside limits raised as a query
It reaches a data queue rather than the investigation procedure.
- LS-05
Cleaner repeat shown as the answer
A later passing result sits above the original as though it settled it.
Stage flags
Readings, gaps and records short of what a result needs
04 · File / handoff
1 mode
- LS-06
Record reads complete when it is not
Raw data missing behind a result does not show in the file.
Stage returns
The sample file an analyst and a supervisor read
05 · Change / Version
1 mode
- LS-07
Limit changed outside change control
A threshold moves without the approval the system requires.
Stage tracks
Specification, limit, model, prompt and rule changes
- Sev-1 · a result loses the procedure behind it
- Sev-2 · the sample file misstates the sample
- Sev-3 · flags degrade, the laboratory re-checks
Affected slices
A tube that changed hands is not a tube that never left
Supervisors drop or rewrite flags before the file is worked, and that is what this number counts. Paper custody has to be transcribed before it can be checked, and a frozen shipment arrives with somebody else's history.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Samples received from a CMO | 5.7% | 3.7× | Review | |
| Hand-written and scanned custody | 4.4% | 2.9× | Review | |
| Cold-chain shipments with an alarm | 2.9% | 1.9× | Watch | |
| Routine in-house release samples | 1.2% | 0.8× | Normal |
Bar: dropped-flag lift vs. in-house release baseline · scale 0–4.0× · tick at the 2.0× threshold
2 of 4 slices over threshold
Evidence-linked improvement
Nothing is fixed at the bench and left there
A flag the supervisor rewrote points at a limit, a custody rule or a protocol map. That is where the fix goes, under change control, before the next run.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
A cohort starts costing more re-checks than it used to.
- 02
Diagnose
The cohort is the clue — samples in it share a route into the lab.
- 03
Improve
The change clears your change control before the agent uses it.
- 04
Verify
Affected samples are read again from login, the failures included.
- 05
Learn
The sample that failed is kept, and later releases run against it.
Learn → Detect
The return edge. A limit or a rule the agent checks against lives in a validated system — it moves with a named approver and a record, not quietly between runs.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, access and login, custody and conditions, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and boundary definition.Week 1
- 02LIMS, instrument and chamber access review.Week 1, Week 2
- 03Specification, limit and protocol mapping.Week 1, Week 2
- 04Sample login and identity checks.Week 2
- 05Custody, location and aliquot parentage tracking.Week 2, Week 3
- 06Storage, excursion and stability-timepoint comparison.Week 3
- 07Turnaround and test-assignment checks.Week 3, Week 4
- 08Record-completeness checks behind a result.Week 3, Week 4
- 09Investigation routing to a named person.Week 4
- 10Analyst, supervisor and QA review workflow.Week 4, Week 5
- 11Evaluation suite and regression samples.Week 4, Week 5
- 12Integration, deployment and Agent Care handover.Week 5, Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne sample type, one site | ProductionProduction LIMS integration | AdvancedSeveral sites / systems |
|---|---|---|---|
| Introduced at Pilot | |||
| Login, identity and custody checks | ✓ | ✓ | ✓ |
| Storage, excursion and timepoint comparison | ✓ | ✓ | ✓ |
| Investigation routing to a named person | ✓ | ✓ | ✓ |
| Attributable trail written for review | ✓ | ✓ | ✓ |
| Release left to a supervisor and QA | ✓ | ✓ | ✓ |
| Record-completeness checks | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| LIMS, instrument and chamber integration | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Several sites, product families and systems | — | — | ✓ |
| Multi-site and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on the sites and systems in scope, sample types and volume, storage and stability coverage, review controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your specifications, limits and stability protocols
→Specification, limit and protocol mapping
Week 102How a sample is logged, labelled and handed over today
→Sample login and identity checks
Week 203Read access to the LIMS, chamber logs and instrument data
→LIMS, instrument and chamber access review
Week 204Custody records as they really are, paper included
→Custody, location and aliquot parentage tracking
Week 305Your out-of-specification procedure, as your SOP writes it
→Investigation routing to a named person
Week 406Samples that went wrong — a mix-up, a missed excursion
→Evaluation suite, regression samples and failure-mode testing
Week 407Named analysts, QC supervisors and a QA reviewer
→Review workflow, then pilot samples and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. Week 3 is where custody and storage meet, and the later checks rest on both.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Specifications, limits, protocols and the automation boundary
W2LIMS, chamber and instrument access, and sample login checks
W3Custody, aliquot parentage, storage and timepoint comparison
W4Investigation routing, record checks and sample cohort slices
W5LIMS integration, first supervisor-reviewed sample files
W6Production validation with your analysts, then Agent Care handover
- Reading the band
- Bars cover only the weeks their work is named in. Nothing runs on a live sample before week 4, which is when routing is put against your own out-of-specification procedure.
- At the end of W6
- A supervisor has worked a real sample file and said where it was wrong. Limits and the version map pass to Agent Care after that.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Biotechnology AI agent
Build a sample-tracking agent around the LIMS you already work in.
Show us one sample type, the systems its history sits in and how a result outside specification reaches your QC supervisor today. We write down where the agent stops before we write down what it does. An out-of-specification procedure that software is allowed to shorten has stopped being a procedure.
- Typical delivery
- 4–6 weeks
- Deployment
- Your LIMS and laboratory stack
- Result release
- A QC supervisor and QA, always
- After launch
- Agent Care available
Nestack Agents · LIMS & sample tracking
AGT-BT-05 · Agent Care available after launch