PlatformThe CMC platform for biologics

Take the waitingout of CMC.

One governed record for every batch, run and document you own, and AI that investigates, analyzes and drafts on top of it.

Book a demo See where the time goes

Your data stays in your environment

Process development

process development

Process development workflow

Edit workflow

Upstream process development (Mab) workflow

Create upstream development report

Studies

Study 1: Temp and pH shift

10 runs · complete

Study 2: Feed study

8 runs · in progress

Equipment

10 L Sartorius bioreactorBRX-0010-03 · in calibration
Cedex cell counterCDX-4402 · in calibration
Mettler Toledo pH meterPHM-1188 · due in 21 days
Nova Biomedical bioanalyzerNVA-7731 · in calibration

Materials

Basal mediumBM-2401-114
Feed AFEED-M-2402
Base, 1 M Na₂CO₃BS-2312-009
Antifoam CAF-2401-052

Process parameters

Parameter	Target range	Value
Working volume	6 – 10 L	8 L
Seeding density	0.3 – 0.5 ×10⁶/mL	0.4 ×10⁶/mL
Starting pH	6.8 – 7.4	7.3
pH deadband	± 0.03 – 0.08	± 0.05
Temperature shift	32 – 34 °C	34 °C
Temperature shift day	6 – 8 days	7 days
Dissolved oxygen	20 – 50 %	40 %
Agitation	80 – 150 rpm	100 rpm
Air sparging	0.01 – 0.1 vvm	0.02 vvm
Feed start	day 3 – 4	day 3
Feed rate	3 – 5 % v/v daily	3 % daily
Osmolality	280 – 320 mOsm/kg	298 mOsm/kg
Antifoam	0.01 – 0.05 %	0.02 %
Harvest criterion	viability < 80 % or day 14	day 14

In-process controls

Viable cell density, viabilitydaily
Glucose, lactate, glutamine, ammoniadaily
Osmolalityevery other day
Offline pH, pCO₂daily
Titerday 7, 10, 12, harvest

Procedure

written once · executed ten times · two values assigned per run

1Seed at 0.4 ×10⁶ cells/mL, 36.5 °C, pH 7.10
2Feed A at 3 % of working volume daily from day 3
3Shift temperature to 33 °C on the assigned day
4Hold pH at the assigned setpoint from that day
5Harvest at viability below 80 %, or day 14

Run data

recorded against the run · nothing re-keyed

Run	Shift day	pH	Peak VCD	Titer g/L	Viability
Run 1	3	6.90	19.0	3.36	82 %
Run 2	3	7.10	19.7	3.45	81 %
Run 3	4	7.00	20.4	3.74	81 %
Run 4	5	6.95	21.1	3.84	80 %
Run 5	6	7.00	21.8	3.94	80 %
Run 6	6	7.20	22.5	3.59	79 %
Run 7	7	6.90	23.2	3.76	78 %
Run 8	8	7.05	23.9	3.80	78 %
Run 9	9	7.00	24.6	3.62	77 %
Run 10	9	7.15	25.3	3.43	77 %

DoEMaterial & equipmentProcedureChartsAnalysisNotes

Harvest titer over culture duration

titer g/L · culture day on the x axis

Online pH over culture duration

online pH · the shift day is visible per run

Viable cell density and viability

median with the spread across all ten runs · viability dashed

Titer across the two factors the study varied

shift day on the x axis · pH setpoint on the y axis

Ten runs, plotted from the data the operators recorded. Nothing was re-keyed to get here, and the same numbers feed the study report.

DoEMaterial & equipmentProcedureChartsAnalysisNotes

Analysis design tableProcess analyticsRecommendationsModels

HeatmapPCARSMCumulative variance …

Principal component analysis shows how the runs cluster. Draw a region on the plot and the platform turns it into recommended ranges.

What the data says

Shift day and pH setpoint together explain most of the spread in titer10 runs · response surface
The best predicted condition is a day-6 shift with pH held at 7.00Runs 5, 8 · 3.94 and 3.84 g/L
Past day 8 the gain is given back through viability at harvestRuns 9, 10 · viability 77 and 76 %

Recommended next steps

1Three confirmation runs at day 6, pH 7.00The optimum is interpolated, not executed. Nothing moves forward on a predicted point.
2Narrow the pH range to 6.95 – 7.05 for study 2Outside that band the model loses more than 0.2 g/L.
3Carry feed rate in as the next factorIt was held at 3 % throughout, so its effect is unmeasured here.

Awaiting a named scientistA recommendation is a draft until someone accepts it into the next study.Every claim cited

10 L → 2000 L

Study 1 confirmed condition, into the 2000 L train

day-6 shift · pH 7.00 · geometric similarity maintained

volume ×200linear ×5.85

	10 L bench	2000 L production
Working volume	8 L	1600 L
Tank diameter, T	0.22 m	1.29 m
Impeller diameter, D	0.10 m	0.58 m
D/T	0.45	0.45
Impeller	3-blade segment	3-blade segment
Power number, Np	1.27	1.27
Agitation	100 rpm	to be set
P/V	7.3 W/m³	to be set
kLa	14 h⁻¹	≥ 12 h⁻¹ required
Mixing time, θm	12 s	to be set

Three criteria, evaluated against the receiving vessel

P/V = Np ρ N³ D⁵ / V · v_tip = π N D · kLa = A (P/V)^0.5 (v_s)^0.5 · θm ∝ T^(2/3) (P/V)^−1/3

Constant P/V

viable

holds power per unit volume

N: 31 rpm
P/V: 7.3 W/m³
v_tip: 0.95 m/s
kLa: 14.0 h⁻¹
θm: 39 s

Holds the power input, and with it the oxygen transfer.

Constant tip speed

fails

holds impeller tip speed

N: 17 rpm
P/V: 1.2 W/m³
v_tip: 0.52 m/s
kLa: 5.7 h⁻¹
θm: 71 s

Holds shear at the impeller, and gives up most of the power.

Constant kLa

viable

holds oxygen transfer, sparge raised

N: 27 rpm
P/V: 4.5 W/m³
v_tip: 0.81 m/s
kLa: 14.0 h⁻¹
θm: 46 s

Buys the transfer back through gas flow instead of agitation.

The three cannot be held together. Geometric similarity fixes D/T, so P/V, tip speed and mixing time move against each other, and the criterion is a choice about which one the process is most sensitive to. Oxygen demand at peak density needs 12 h⁻¹, which rules out constant tip speed at 5.7 h⁻¹.

Proposed basis: Constant P/V

Agitation

31 rpm

P/V

7.3 W/m³

kLa

14.0 h⁻¹

θm

39 s

Temperature, pH and feed carry across unchanged: they are process values, not scale-dependent ones. Only what the vessel changes has been recomputed.

What it costs, and what it does not

Mixing time rises 12 s → 39 sBase and feed additions see a gradient before they disperse. Move the feed to a dip tube at the impeller and slow the bolus.
pCO₂ accumulates at scaleHeadspace-to-volume falls with the cube root, so stripping is sparge-limited. Hold pCO₂ under 100 mmHg or the pH control fights it.
Tip speed 0.52 → 0.95 m/sBelow the 1.5 m/s working limit for CHO, so no hydrodynamic change is expected.
The kLa constant is not transferableBench uses a microsparger, the 2000 L a drilled-hole sparger. Bubble size and coalescence differ, so A is confirmed by a gassing study on the receiving vessel before the first cell.

Transfer package, drafted from the record

Scale-up rationalethe criterion, the two it was chosen over, and why
Parameter ranges for the 2000 L MBRcomputed, each carrying its correlation
Gassing study protocolto confirm kLa on the receiving vessel
Comparability planwhich attributes are compared against the bench runs

Approved by R. Iyer, MSAT · e-signedThe engineer owns the criterion. The platform did the arithmetic and showed its working.

evaluating against the plan

SpecificityDilutional linearityAccuracyRepeatabilityIntermediate precision

Plate layout · 96 well

123456789101112AA1Std 10004502.2066500.038OD2.168A2Std 10004502.2056500.037OD2.168A3Std 10004502.2076500.039OD2.168A4Smp4500.7866500.038OD0.748A5Smp4500.7616500.037OD0.724A6Smp4500.7806500.039OD0.741A7Ctrl4500.3626500.038OD0.324A8Ctrl4500.3726500.037OD0.335A9Ctrl4500.3696500.039OD0.330A10<LLOQBlk4500.1056500.038OD0.067A11<LLOQBlk4500.1016500.037OD0.064A12<LLOQBlk4500.1056500.039OD0.066BB1Std 3004502.1106500.038OD2.072B2Std 3004502.1096500.037OD2.072B3Std 3004502.1116500.039OD2.072B4Smp4500.7766500.038OD0.738B5Smp4500.7516500.037OD0.714B6Smp4500.7706500.039OD0.731B7Ctrl4500.3526500.038OD0.314B8Ctrl4500.3626500.037OD0.325B9Ctrl4500.3596500.039OD0.320B10<LLOQBlk4500.0956500.038OD0.057B11<LLOQBlk4500.0916500.037OD0.054B12<LLOQBlk4500.0956500.039OD0.056CC1Std 1004501.8286500.038OD1.790C2Std 1004501.8276500.037OD1.790C3Std 1004501.8296500.039OD1.790C4Smp4500.7916500.038OD0.753C5Smp4500.7666500.037OD0.729C6Smp4500.7856500.039OD0.746C7Ctrl4500.3676500.038OD0.329C8Ctrl4500.3776500.037OD0.340C9Ctrl4500.3746500.039OD0.335C10<LLOQBlk4500.1106500.038OD0.072C11<LLOQBlk4500.1066500.037OD0.069C12<LLOQBlk4500.1106500.039OD0.071DD1Std 304501.1636500.038OD1.125D2Std 304501.1626500.037OD1.125D3Std 304501.1646500.039OD1.125D4Smp4500.7716500.038OD0.733D5Smp4500.7466500.037OD0.709D6Smp4500.7656500.039OD0.726D7Ctrl4500.3476500.038OD0.309D8Ctrl4500.3576500.037OD0.320D9Ctrl4500.3546500.039OD0.315D10<LLOQBlk4500.0906500.038OD0.052D11<LLOQBlk4500.0866500.037OD0.049D12<LLOQBlk4500.0906500.039OD0.051EE1Std 104500.5426500.038OD0.504E2Std 104500.5416500.037OD0.504E3Std 104500.5436500.039OD0.504E4Smp4500.7836500.038OD0.745E5Smp4500.7586500.037OD0.721E6Smp4500.7776500.039OD0.738E7Ctrl4500.3596500.038OD0.321E8Ctrl4500.3696500.037OD0.332E9Ctrl4500.3666500.039OD0.327E10<LLOQBlk4500.1026500.038OD0.064E11<LLOQBlk4500.0986500.037OD0.061E12<LLOQBlk4500.1026500.039OD0.063FF1Std 34500.2166500.038OD0.178F2Std 34500.2156500.037OD0.178F3Std 34500.2176500.039OD0.178F4Smp4500.7686500.038OD0.730F5Smp4500.7436500.037OD0.706F6Smp4500.7626500.039OD0.723F7Ctrl4500.3446500.038OD0.306F8Ctrl4500.3546500.037OD0.317F9Ctrl4500.3516500.039OD0.312F10<LLOQBlk4500.0876500.038OD0.049F11<LLOQBlk4500.0836500.037OD0.046F12<LLOQBlk4500.0876500.039OD0.048GG1Std 14500.1246500.038OD0.086G2Std 14500.1236500.037OD0.086G3Std 14500.1256500.039OD0.086G4Smp4500.7886500.038OD0.750G5Smp4500.7636500.037OD0.726G6Smp4500.7826500.039OD0.743G7Ctrl4500.3646500.038OD0.326G8Ctrl4500.3746500.037OD0.337G9Ctrl4500.3716500.039OD0.332G10<LLOQBlk4500.1076500.038OD0.069G11<LLOQBlk4500.1036500.037OD0.066G12<LLOQBlk4500.1076500.039OD0.068HH1<LLOQStd 0.34500.0976500.038OD0.059H2<LLOQStd 0.34500.0966500.037OD0.059H3<LLOQStd 0.34500.0986500.039OD0.059H4Smp4500.7786500.038OD0.740H5Smp4500.7536500.037OD0.716H6Smp4500.7726500.039OD0.733H7Ctrl4500.3546500.038OD0.316H8Ctrl4500.3646500.037OD0.327H9Ctrl4500.3616500.039OD0.322H10<LLOQBlk4500.0976500.038OD0.059H11<LLOQBlk4500.0936500.037OD0.056H12<LLOQBlk4500.0976500.039OD0.058

Standard curve Signal (OD)

7 standards on the fit48 sample and control reads✕25 below the quantifiable range

4PL model

y=d+a − d1 + (x/c)^b

SpecificityPass

LinearityPass

AccuracyPass

RangePass

Limit of quantitationPass

RepeatabilityPass

Intermediate precisionPass

live

Open deviations

2 with a proposed cause

Batches in review

avg 9 s to first pass

Median days to close a deviation

Every batch in review is reconciled before a reviewer opens it. The next one down the list is batch 381.

Batch Record 381, 2000 L production bioreactor

MBR v4 · executed 11–24 Jun · 128 signed steps · 41 pages

Reviewed in …

Parameters vs recipe

Recipe ranges × recorded values

Equipment calibration

Equipment historian

Operator training

Training system

Material & consumable lots

Inventory · QC status, expiry

Exceptions: the only pages a reviewer needs to open

PARAMDay 6 pH held at 7.19, recipe range 6.95–7.15Step 6.2 · 4 h 20 min above range
PARAMHarvest titer 3.10 g/L against a 3.50 g/L lower limitRelease QA · deviation already raised
EQUIPDO probe DO-4471 calibration expired 3 days into the runBRX-2000-07 · historian record
MATERIALFeed lot FEED-M-2402 released, osmolality outside historical bandCoA on file · no restriction flag

The other 124 steps reconciled clean against the master batch record. The titer exception already carries a raised deviation, so the investigation opens with the evidence attached.

Harvest titer · last 8 batches

Batch 381 came in at 3.10 g/L

Raised at the point of record

Evidence gathered automatically · read-only, your data only

7 comparable runs, same recipe and scale
3.49 – 3.62 g/L · this run 3.10
Run data
Bioreactor BRX-2000-07 calibration
In calibration, verified 12 days before the batch
Equipment historian
Operator and verifier training
Both current on SOP-USP-014 at execution
Training system
Feed media lot FEED-M-2402
Released, osmolality 318 vs 295–305 mOsm/kg historical
Inventory / CoA

Proposed cause, 6M category: Material

Feed media lot FEED-M-2402 ran 13 mOsm/kg above the historical band for this recipe.

A genuine alternative stays on the record rather than being argued away, Environment, the day-4 room excursion, overlaps the same window.

Awaiting a named reviewerEvery claim cited · full trail retained

The operating reality

Nobody is short of expertise.
They're short of access to it.

These costs show up in every biologics organization we have worked with, from the first development run to the filing. None of them are science problems.

01/ 05Process development

Studies are designed, run and analyzed in three different tools.

Design, data and conclusion end up in separate files, so the reasoning behind a set point has to be reconstructed years later.

02/ 05Analytical

A result gets checked against a limit, not against its history.

Out of specification is the easy call. Out of trend needs the method's own past runs, and those sit in files nobody opens.

03/ 05Scale-up

Scale-up rests on who is still in the building.

What the sending site did, and why, sits in memory and inboxes rather than in the record that moved with the process.

04/ 05MSAT

Batch review is page-turning, not judgment.

MSAT reads every signed step by eye, then checks equipment, training and lot status in three other systems, to find the handful of lines that matter.

05/ 05Reports

Reports are written by re-typing data you already have.

Senior scientists assemble reports by copying numbers out of spreadsheets: the most expensive typing in the industry.

The product

Wherever you are in the process,
these three are always there.

Your process history, the chat over it, and the documents that come out. The domain work has a page of its own: qualifying a method, reviewing a batch, running an investigation, designing a study.

Contextualized process knowledge

Years of process history, answerable in a sentence.

Reports, runsheets and batch records are split into sections and linked to the study or batch they describe. Every answer arrives with its page attached.

Tech transfer · MSAT · new starters · audits

indexing

Sources · 318 pages

BPAI001 Upstream Development Report.pdfDevelopment report·84p
Bioreactor Study 1, Temp & pH shift.xlsxRun data·12p
Batch Record 381 (filled).pdfBatch record·41p
HCP ELISA Qualification Report.pdfAnalytical·63p
BPAI001 Tech Transfer Document.pdfTech transfer·118p

ASKWhy did we shift temperature to 33 °C on day 6, and which batches ran it?

The shift was introduced in Study 1 to hold viability above 80 % through harvest after the day-6 viability drop. It carried into batches 251, 381 and 398; batch 405 at the CDMO did not adopt it.

Upstream Dev Report · p.41Study 1 runsheet · rows 18–24Batch 381 · step 6.2

One chat, wherever you are working

Any scientist can interrogate a campaign, without a data scientist in the loop.

Ask in plain language. The model chooses the analysis and writes the interpretation; the platform produces every number.

Process development · MSAT · Analytical

ready

Across our last 8 production batches, what actually drives harvest titer? Plot it and tell me whether the day-6 temperature shift is earning its place.

Found the run data for all 8 batchesdone

Pulled titer and peak cell densitydone

Compared every variable against every otherdone

Fitted titer against cell densitydone

Harvest titer vs peak VCD

g/L against ×10⁶ cells/mL · n = 8

Peak viable cell density explains most of the variation in titer (R² 0.90). The day-6 temperature shift correlates positively but weakly (r 0.44), so on this dataset it is not doing the work you think it is. Batch 381 sits off the line on both axes.

Every number here was calculated over your own data. The AI chose which analysis to run and how to describe it. It never made a value up.

Reports & submission content

Report writing becomes a draft your author edits.

Structure comes from your approved template, held as data. Figures are recomputed on every regeneration: the model writes the prose between the numbers, never the numbers.

PD leadership · Regulatory · CDMO client delivery

assembling

Report template · v3 (approved)

1. Purpose & scopeTemplate
2. Process descriptionRecipe of record
3. Materials & equipmentInventory · catalog
4. Results & discussionRun data · 8 batches
5. Deviations & investigationsInvestigation records
6. ConclusionDrafted · needs review
Appendix A, raw dataResult tables

Structure is data, not code. Your template, section order, headings, house language, governs every document of this type, for every molecule.

BPAI-001 · Upstream Development Report · Draft v0.1

4. Results and discussion

Eight production runs were executed against MBR v4 between June 2024 and April 2026. Harvest titer averaged 3.51 g/L with a standard deviation of 0.17 g/L. Seven of eight runs met the 3.50 g/L lower limit. Peak viable cell density correlated strongly with titer across the campaign (R² = 0.90).

One run, Batch 381, fell below the limit at 3.10 g/L. The investigation is recorded in section 5 and concluded on feed media lot FEED-M-2402.

Every highlighted value is bound to the record

Numbers are read from the result tables and recomputed on regeneration, the model writes the prose around them, never the figures. Change the data, the document changes with it.

What changes

Your team keeps the science.
The work around it goes away.

None of this comes from running the experiments faster. It comes from the assembly work between an experiment and a decision, which is where the weeks actually go.

01Analytical · PD authors

The report arrives as a draft, not a blank page

A qualification or development report comes assembled from the study that was actually executed, figures already in place and recomputed if the data changes. Your author edits and signs instead of transcribing.

02MSAT · Quality

Batch review stops being page-turning

You see the handful of steps that deviated instead of reading every signed line, with equipment status, training and lot status already checked against the record.

03MSAT · Manufacturing

An investigation opens with the evidence already gathered

The runs, the trend against the method's own history and the comparable batches are pulled together before the first meeting, not during the third.

04Tech transfer · Audits · New starters

A question about a campaign from three years ago takes minutes

Answered from the record, with the page attached, without having to find the person who ran it or hope they wrote it down.

Trust and security

Can I trust the answer,
and can I trust you with the data?

The two questions every serious evaluation comes down to. Three answers, and a page for the people whose job it is to check them.

Trust the answer

Every claim carries the record it rests on, down to the page it was read from
Where the evidence is not there it says so, rather than filling the gap with something that sounds right
Anything that would become part of the record stops for a named reviewer and an electronic signature

Protect the data

Your record sits in a database of your own and is never pooled with anyone else's
Encrypted in transit and at rest, with access controlled through your own directory
Every question, step and approval recorded against the record it touched

Enterprise security

Runs on Microsoft Azure, on infrastructure audited to ISO 27001 and SOC 2
Cross-region replication and failover for the application and the data
Built for 21 CFR Part 11 and EU Annex 11: attributable, contemporaneous, inspectable

Security and data handling, in full

SOC 2 Type IIIndependently audited
ISO 27001Certified
21 CFR Part 11Built to comply

Portfolio coverage

A CDMO's portfolio is mixed.
Your platform should be too.

The record is modelled on unit operations, parameters and results, not on one molecule class. A mAb, an AAV and an mRNA process sit on the same system.

Bench & development0.25 – 50 L

Pilot & clinical50 – 500 L

Commercial500 – 20,000 L

Modalities in production use or supported

Monoclonal antibodiesBispecifics & multi-specificsAntibody fragmentsAntibody-drug conjugatesRecombinant proteinsCell therapyGene therapy (AAV, LV)mRNA & LNPOligonucleotidesVaccinesMicrobial & fermentationPlasmid DNA

Upstream and downstream, drug substance and drug product, in-house sites and partner CDMOs, all in one place.

From the people who do the work

What practitioners told us.

Anonymized feedback from process development, manufacturing and CDMO leaders.

As a biotechnology professional with over 20 years of experience, I found this to be really helpful in process development, tech transfer, scale up and manufacturing activities. It also significantly reduces the time taken to draft technical reports and regulatory documents

Executive Director, Mid Size Biotech Company

Being a CDMO, this capability is very attractive.

Sr Director, US based CDMO company

BioprocessAI serves as a comprehensive solution for all our process development needs. It covers all aspects from experimental design guidance, execution, data visualization, analysis, to report and regulatory document writing. Whole package!

Head of Downstream Process Development, European CDMO company

This AI based solution for Bioprocessing is an excellent platform and I see an immense potential in developing it, catering to the bioprocessing community and manufacturers

Manager, Biotech Company

Bring us the work you
already know the answer to.

A closed investigation. A released batch. A report your team wrote last quarter. We will run it on the platform, and you judge the output against work you can already check.

Book a demo Talk to us

or email sales@bioprocess.ai