Manufacturing that learns.

1A traditional MES runs the recipe. Seal improves the next batch.

A traditional MES executes the approved recipe and records what happened. It guides operators through the steps and files the batch record, but the laboratory result, the deviation and the training record sit in other systems, and once the configured process is validated it changes as rarely as possible.

So an investigation that ends with “add a check before the hold” leaves the work unfinished. Someone must carry the change from the quality system into the master batch record, verify it, train the operators and establish which version is now in force.

Seal is one platform for manufacturing execution, laboratory, quality, training and inventory. Build the manufacturing system around your process: material eligibility, equipment readiness, operator instructions, QC and review belong to the same workflow, so the batch record grows as the work happens. A corrective action can be released as a new version of the process itself, carrying its specifications, test evidence and approval.¹ The next batch runs the improved process; earlier batches keep the version they ran.

You can start with one process, alongside the MES, LIMS or QMS you already run. The advantage grows as manufacturing, laboratory and quality share the foundation: an investigation can lead to a new check at the affected step, and a transferred process can become the starting point for another site. The platform supports both today’s production and the operation you are building next.

2One platform for the whole batch.

Manufacturing execution, LIMS, electronic lab notebooks, quality, training and inventory share one configurable platform in Seal, so a batch links to every record it depends on.

A lab result belongs to the batch step it governs. A deviation opens with the affected material, equipment and procedure in view. The operator’s qualification is available where the work requires it.

Forms, calculations, dependencies and review gates define how the work runs. Permissions, signatures, audit history and process versions are shared controls, and the specifications and verification evidence stay with the configuration as it changes. When a process runs in Seal, its instructions, measurements, exceptions and approvals become connected records, so the next investigation starts with the history of the operation, including what changed and why. The Seal platform describes the shared foundation in full.

You can begin without replacing the systems you rely on. Where an existing MES, LIMS, QMS or ERP remains in place, connect the records the job needs and agree which system owns them.

Each connection is scoped before it goes live: what Seal may read, whether anything is written back, which system stays the source of record and how the mapping is verified. Historian and control-system connections read values; they do not change setpoints, acknowledge alarms or operate equipment. Some connections are set up per project on the vendor’s API. See all integrations and the supported instruments.

3Already running in production.

MiAlgae brought its electronic lab notebooks, quality management and batch records into Seal as it scaled from lab to commercial production. Its customer story reports a 75% reduction in documentation time. Operators follow guided workflows; measurements and quality events stay linked to the batch. Read the MiAlgae story.

At Skye Biologics, the same platform covers batch records, document control, training, inventory and instrument integration: the process, its source data and its review in one place. Read the Skye Biologics story.

4Neil prepares the work. Your team decides.

Neil is Seal’s AI agent. It works across the same manufacturing, laboratory and quality records as your team, within the permissions of the person asking, and what it produces is work in Seal: a configured process, an answer with its sources or a prepared change.

Everything in the middle column is preparation. Nothing Neil produces takes effect until a qualified person has reviewed it, and operators run only the approved version.

In line with the EU’s draft GMP Annex 22 on AI, Neil is not used in GMP execution. It helps set up configuration, which your team verifies and releases under change control. Neil follows the requester’s access permissions, and customer data is not used to train AI models. See how Neil fits Annex 22, Neil at work and the AI and data controls.

4.1Configure the process

Give Neil a procedure or master batch record. It proposes the steps, fields, calculations, material and equipment requirements, and review gates in Seal. Process experts check the source mapping and resolve assumptions before testing and approval. Operators run the approved version.

4.2Answer with the source records

Ask “Which batches used material lot RM-208?” or “Which procedure version governed this run?” Neil follows accessible manufacturing, lab and quality records, including records in connected systems, and answers with sources the team can inspect.

4.3Investigate and prepare the next change

Ask what changed between runs. Neil compares timestamps, procedure versions, results and linked deviations, then proposes an investigation or process change. The finding, its sources and the proposed action stay connected for review.

In this example, the step 07 hold has lengthened over four batches, and only B-041 is over the 45-minute limit. The next question is whether the receiving team was ready when each hold began. The proposed readiness check becomes a change to assess and test. The limit stays in place, and the source records do not establish a cause. Looking across runs is what GMP expects of an investigation: it extends to other batches that may be associated with the discrepancy.²

5What happens during a run.

The approved workflow checks each prerequisite before a step starts: procedure version, operator qualifications, material and equipment status, and line clearance. It also defines what happens when one fails: block the step, require verification or follow an exception path.

Material is checked before it enters the batch. The operator scans the lot, the workflow checks its identity, status and expiry, and the quantity is recorded against the configured tolerance. The accepted dispense links the lot and actual quantity to the operation, which is the source for inventory reconciliation and batch genealogy. Each step keeps what GMP asks the batch record to show: the date, the equipment, the component lot, the weights and measures, and who performed and checked the work.³

Operators follow the current instruction, scan material lots and record measurements from supported instruments. Calculations use the configured expressions and units. These are controls in the workflow: execution does not depend on Neil interpreting the instruction afresh at every step.

A captured reading keeps its value, device identity and source timestamp together. Where a value is entered manually, configure the required verification and retain that entry’s provenance too. Capturing data from an instrument and controlling it directly are separate scopes, each agreed and verified on its own.

6Review results and exceptions in context.

Review by exception brings forward what needs judgement: a failed check, an out-of-range result, a missing approval or a documented departure from the procedure. The full execution history stays available alongside it.

DEV-018 carries the original timestamps and the applicable requirement into review. The authorised reviewer assesses the departure, records the disposition and retains the evidence. The quality unit reviews and approves the production records before a batch is released;² review by exception directs that review to the entries that need judgement without hiding the rest of the batch.

An in-process test belongs to a specific batch and operation. Seal links the sample, method, specification and result, and the workflow can make the next operation depend on that result. Here, pH 7.10 meets the example range, but step 13 stays blocked until the review is recorded.

An acceptable number and an authorised decision are distinct records. The records and limits here are examples; the approved process defines the real requirements. The laboratory and quality management blueprints describe those workflows in full.

7Trace materials and see what is ready.

Raw-material lots, intermediates and finished units link to the operations that use or produce them, so a trace runs in both directions: from a finished batch back to its inputs, or from a material lot forward to the batches that used it.

Equipment, operators and test results stay with the trace as context. Agree the identifiers, units and source of record when configuring the process, particularly where inventory is also maintained in an ERP.

The same records drive facility status. Every status has a record behind it: the current operation, the pending test, the deviation under review or the clearance needed for the next run.

8Change the process without losing its history.

A finding becomes a new version of the process, carrying its requirements, tests and approval. Open runs keep the approved version they started on.

An investigation becomes useful when its conclusion changes how the work runs. Neil can propose a revised configuration from the procedure and supporting evidence, and the team reviews the differences against the current version.

Continuous validation is integrated into the platform. User requirements (URS), functional and design specifications (FS and DS), the configuration specification (CS) and automated verification and UAT evidence stay linked to the process version, rather than living in a separate handover pack.⁴ For the readiness check, that means testing the blocked path when acknowledgement is missing and the permitted path when it is recorded, then retaining the results, exceptions and approval with the new version.¹

Intended use and risk determine the testing scope, including affected interfaces, calculations, permissions and exception paths. Approval authorises the new process; subsequent runs provide the evidence to assess whether it worked. The continuous validation detail explains the risk-based approach and who is responsible for what.

9Work on the questions that matter.

Connected execution is the foundation for deeper process understanding. Our long-term ambition is to put increasingly capable AI to work on yield, scale-up and patient-specific production. Today the work starts with the evidence available, identifies what is missing and turns promising hypotheses into experiments the team can evaluate.

9.1Explaining a change in yield

A lower yield might coincide with a different material lot, an extended hold, an equipment change or a revised method. Establish which runs are comparable, identify missing measurements and separate observations from possible explanations. The useful output is the runs to examine, the evidence for each hypothesis and the next experiment that could distinguish between them. Finding a pattern is the beginning of that work, not proof of its cause.

9.2Carrying context through scale-up

Carry development experiments and their rationale into the manufacturing record, so the receiving team inherits the context behind a parameter rather than a number alone. Process scientists still decide what carries across, what needs a new model or experiment, and what must be demonstrated on the receiving equipment.

9.3Recognising a recurring exception

A closed deviation should leave more than a completed form. Keep the finding, competing explanations, corrective action and effectiveness evidence with the affected process, so a later investigation can establish what was tried and what remains unresolved.

9.4Manufacturing for one person

Patient-specific production makes identity, material, process and disposition central to each run: chain of identity and custody, time-sensitive handoffs and the evidence required before release. A shared platform gives teams a place to configure those relationships, though therapy-specific workflows, interfaces and identity checks still need to be scoped, verified and approved for their intended use.

9.5A medicine made for every person

That is our mission. A medicine must be made reliably, tested and delivered, and for a therapy made from one person’s cells the operational system must carry the identity, process and proof of that individual run. Seal provides the shared operational foundation and Neil puts the available context to work, carrying knowledge forward from development to manufacturing, between sites and from one dose to the next.

10Start with one job. Expand on the same platform.

10.1Choose the work

Bring a recurring exception, a batch question or a representative procedure. Together we agree the question to answer or workflow to run, the records needed and what a useful result looks like. Use sample material or arrange an NDA before sharing confidential records.

10.2Inspect the result

For an investigation, examine Neil’s findings against the cited sources. Once we agree the process evaluation scope and receive the required inputs, see the first working configuration and supporting validation pack within 48 hours. Run the workflow, try its exception paths and inspect the requirements, test results and configuration history.

10.3Decide what goes into production

Manufacturing, quality and IT review the process, interfaces, migration, device use, training and support needed for release. The evaluation is a way to make that decision with evidence, not automatic production approval. Expand into linked operations as the scope is approved, keeping the systems you still need.