Manufacturing that evolves with your therapy.

1A traditional MES records the batch. Seal carries the patient through it.

A traditional MES was built for a process that runs the same way every time, on material that arrives as specified. In cell and gene therapy, chain of identity often sits in a labelling system that is not connected to the manufacturing floor, the batch record assumes starting material that patient cells rarely provide, and review begins when manufacture ends. Each handoff between separate systems is a gap in the data.

Seal is where the work runs. The product, the patient or donor identity, material genealogy, process versions, actual work and release evidence belong to the same model. The identity check is part of each step, the batch record follows what the cells do, and release evidence is assembled while the batch runs.

A therapy’s manufacturing operation must develop alongside the science. Starting material, analytical methods, equipment and the process itself can change between early studies and routine production. In Seal, each change becomes the next version of the workflow, assessed for its intended use and released under change control, while the lots already made keep the basis they were made on.

2Neil prepares the change. Your team decides.

Neil is Seal’s AI agent. It works across the same records your team does, within the permissions of the person asking, and what it produces is work in the system: a receipt check, an exception path, a process revision or an assembled release package. It does not hand back text to paste somewhere else.

Give Neil the receipt requirements and it prepares the identity checks, the exception paths and the cases your team needs to verify: a container that belongs to another identity, a missing collection time, a new version taking effect. Everything it produces is preparation. Nothing takes effect until a qualified person has reviewed it, and lots already in progress keep the version they started with.

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. Customer data is not used to train AI models, and the requester’s permissions govern what Neil can read. See how Neil fits Annex 22 and more about Neil.

3Make chain of identity the data model.

Chain of identity is often handled by a labelling system that is not connected to the manufacturing floor: it generates barcodes and relies on someone scanning them. Seal makes chain of identity part of the record structure itself.

Each relevant material record carries a link to the configured patient or donor identity. The apheresis bag records the patient it came from. The cryovial records the patient it belongs to. The batch record records whose cells it is processing. When a technician scans a vial, Seal records the scan and checks the identity against the batch.

That check applies at each step, transfer and handoff. The chain is enforced as the work runs, and documented as a result.

4Block a mismatch rather than warn about it.

A warning dialog relies on the person reading it. Configure a mismatch as a stop instead: when a scanned container does not match the current batch’s patient, the step cannot proceed.

Under time pressure, people click through warnings they have seen many times. Across thousands of patient batches, a small override rate becomes a real risk.

The exception path requires supervisor involvement and deviation documentation, with the mismatch logged, investigated and resolved before processing resumes. An unreadable label stops the step in the same way, until the material is resolved against its authorised source record. A matching scan satisfies only the identity check; material status, timing and the remaining receipt checks still govern progression.

Configured checks at each handoff stop material being released to the wrong batch, and product being shipped against the wrong patient. Each point where a mix-up could occur can have a gate that prevents progression with mismatched identities.

5Model the product, the material and the patient separately.

The ontology needs to represent more than a patient identifier on a batch. Define the product, process version, patient or donor reference, collection, incoming material, manufacturing lot, intermediate, sample and final container, and link each transformation and custody event to the objects it affects.

That gives the team a route from a released unit to its starting material, processing history and evidence. A platform that models only one patient and one batch cannot describe every cell or gene therapy operation, so configure the model around the product rather than forcing these relationships into a single genealogy. A split creates identified descendants; a pool records all contributing inputs; a sample keeps its source and collection point. Record quantities and dispositions so material use can be reconciled through the process.

Autologous workflows are patient-centric: the patient ID organises the record from collection to infusion, including repeated collections, split material and multiple containers. Allogeneic workflows are donor-centric: lots are tracked from donation through banks and manufacturing, with genealogy connecting donor material to every patient who receives product derived from it. Chain of identity, dynamic batch records and concurrent release apply to both; what differs is how identity is tracked and how genealogy is structured. The autologous, allogeneic and viral vector blueprints describe each model in detail.

Use coded identity references in operational views, with access to identifying information limited to the roles that need it. US tracking rules for HCT/Ps require a distinct identification code that relates the product to its donor and its records, and the code may not include the person’s name or medical record number.¹ A courier, laboratory analyst and clinical coordinator have different jobs. Configure what each can read or change, including documents, labels and linked records, and carry the access design through to exports and integrations.

6Let the batch record respond to the material.

Patient cells vary in quality, quantity and behaviour. Seal batch records branch on recorded measurements, so the record follows what the material needs.

A process tuned for ideal starting material has to accommodate the variation of real patient material. A static batch record assumes consistent input, and deviations from the plan end up as handwritten corrections that a reviewer must interpret later.

A low cell count can route the batch to an expanded culture protocol. Viability below threshold can alert QA and open an investigation workflow. When a step runs long, time-limit tracking shows how much margin remains.

The record reflects how the batch actually ran, including the branches the material required.

7Review the batch while it runs.

A product whose shelf life is measured in hours or days cannot wait for a review measured in weeks. In Seal, QA reviews data during the batch, and the release evidence is assembled as each result is accepted.

The sequence of manufacture, then review, then release does not fit a product made of living cells. EU GMP for advanced therapies recognises that some products must be released before all quality control tests are complete, with certification organised in stages.² Test results are reviewed as they arrive and deviations are investigated as they occur. The Certificate of Analysis is assembled incrementally as each test is accepted.

The authorised reviewer assesses the complete product-specific release package, including identity, required results, deviations and unresolved conditions. The release decision stays with the quality unit; the evidence can be assembled as the work progresses.

7.1Preserve the state of every release requirement.

Configure the product’s review plan around manufacturing completion, identity reconciliation, analytical results, deviations, required assessments and the authorised release decision. Record the source and review state for each requirement. A result can be available, awaiting review, accepted, invalidated or superseded; those states should remain distinguishable.

Where the applicable product strategy permits a result to arrive after an earlier release stage, retain the rationale, required decision and follow-up work in the record. Use the product’s approved arrangements to define the workflow. A dashboard showing completed tasks cannot supply that authorisation by itself.

Connect the release decision to the specific containers and label versions in scope. Dispatch, custody and receipt then extend the same history. A late result or subsequent quality event can be traced to the lots and distributions it affects, with follow-up assigned and its outcome retained. The EU guidelines also expect a procedure for a result that is out of specification after release, including liaison with clinical staff.²

8Run the handoffs from collection to infusion on one record.

Cell and gene therapy involves several organisations: the clinical site that performs collection, the logistics provider that manages cryogenic transport, the manufacturing facility that processes the cells and, often, a different clinical site that performs infusion. Seal gives them one view from collection to infusion, each scoped to its own part.

Each handoff between separate systems is a gap in the data. In Seal, clinical sites are given scoped access to the same record, where authorised users can coordinate collections and the required information. They receive shipping instructions and tracking information, and they can see when their patient’s product enters manufacturing, progresses through processing, passes QC testing and ships back. When a patient asks where their treatment is, the clinical team can answer from the current record instead of waiting on the manufacturer.

Logistics partners see pickup and delivery requirements. Manufacturing has the context for incoming material before it arrives. Access is scoped so that each site works with its own patients’ records, and the status each party reads comes from the record the others write to. Patient orchestration adds clinic enrolment and collection scheduling on the same platform, and a patient hub connection can exchange selected therapy milestones with a patient-support provider.

A manufacturing slot depends on more than a calendar entry. Collection, transport, receipt, equipment readiness, staff qualification, room availability and the selected process all affect whether work can begin. Represent the dependencies and responsible parties together, with actual status and the source time behind it.

Configure handoffs to record the expected container, identity reference, custody, condition and receipt assessment. An expected arrival, a courier delivery and a manufacturing acceptance are different events. Keep delays, damaged containers, missing records and temperature questions connected to the affected material and the decision about its use.

Within manufacturing, record the event from which each hold or processing clock starts, the applicable time zone and the limit configured for that process version. A rescheduled step needs an assessed plan; changing a calendar entry does not change the underlying time allowance. Where external systems provide events, preserve their references and identify missing or inconsistent timestamps.

These relationships let the operations team see what needs intervention across concurrent lots. They also give a clinical coordinator a useful status for the patient’s product, within the information and actions agreed for that role.

9Carry development knowledge into the manufacturing process.

Development teams need freedom to investigate starting-material variability, culture conditions, recovery, potency and the performance of analytical methods. Configure structured protocols that retain actual conditions, interventions, reagent lots and sample lineage while allowing the study to record departures from its plan.

Link a selected parameter or operating range to the studies and observations behind it. Distinguish a result obtained with one donor, scale or analytical method from a conclusion intended to apply more broadly. When a method, reagent or manufacturing site changes, the team can identify which evidence needs reassessment before carrying the conclusion forward.

Neil can compare the permitted runs, surface missing context and prepare the next study or process draft. The same process definition becomes a starting point for manufacturing, with the required identity checks, sampling, calculations and review gates added for its intended use. Earlier studies keep their original versions; the manufacturing version records what has been established and what remains to be resolved. The process development and technology transfer blueprints follow that route in detail.

10Improve the process while protecting patient records.

Repeated manufacturing work should improve the organisation’s ability to deliver. A recurring custody discrepancy may reveal a missing receipt check. An investigation may show that a timestamp needs a clearer definition. A review may expose a calculation that should be built into the workflow. Capture those lessons as proposed improvements to the process components your team maintains.

Neil can help turn the finding into configuration: a required reference, a revised calculation, an exception path or a review step. Persistent memory can retain relevant context and previous decisions; versioned skills can describe how to prepare a lineage assessment, compare a process revision or assemble release evidence. Specialists can refine these methods as they learn from reviewed work.

Keep patient information, donor characteristics, sequences, product-specific limits and restricted programme results within their authorised scope. Reusable skills should contain methods and approved examples appropriate to their audience. Review remembered content and supporting files before reuse across programmes. A generalised lesson requires scrutiny; removing a patient’s name does not establish that the remaining information can be shared.

For example, a team can reuse a method for verifying the lineage of a pooled intermediate while retaining each programme’s identities and measurements in its own records. The skill defines the relationships and checks to inspect; the agent reads the permitted evidence for the current job. This gives the next programme a better method without requiring a combined pool of restricted records.

11Qualify changes against the way the work can fail.

Use a representative product flow to test the configured system: a mismatched container, an unexpected split, an incomplete custody event, a late analytical result and an operator whose qualification does not cover the activity. Verify the intended response and the evidence left for review. Include label reprints, corrected identifiers and integration interruptions where they are part of the operation.

A proposed process revision should identify the affected instructions, fields, calculations, review paths and integrations. Record requirements, test results, deviations and approval with that revision. Existing lots keep the process version they used, while later work can adopt an approved improvement.

Start with one lineage and one manufacturing workflow, then extend the tested components across products and sites. Evaluate whether the team can reconstruct a material’s history, locate an unresolved release requirement and configure the next workflow with less repeated work. Those are concrete demonstrations of a platform that can evolve with the therapy.

12Show the inspector the control, not only the procedure.

Regulators examine cell and gene therapy manufacturing closely. For HCT/Ps within the scope of 21 CFR 1271.290, US establishments must maintain tracking from donor to consignee or final disposition, and back.¹ EU GMP for advanced therapy medicinal products expects a system that tracks cells and tissues in both directions, from donation through manufacturing to delivery to the recipient, with the traceability data retained for at least 30 years after expiry and available rapidly in case of an adverse reaction.³ Map the product’s applicable requirements into its identity, genealogy and review controls.

Seal captures each scan, validation and decision with its context. When an inspector asks how chain of identity is ensured, the answer can include the configured control that enforces it at each step, alongside the procedure that describes it.

Audit trails and electronic signatures are part of the record rather than added afterwards.⁴ They support the manufacturer’s compliance with the applicable requirements; they do not establish it by being present. Documentation is still required, but much of the evidence is captured as the work is done.