The problem we kept seeing
A biologic batch is among the most valuable things a manufacturer will ever make. Weeks of work, a living culture, and for a cell or gene therapy, one patient’s only dose. The process is monitored continuously — pH, dissolved oxygen, temperature, biomass, feed rates, off-gas. There is no shortage of data.
And yet the industry largely treats a run as if it will tell you when it’s failing. It won’t, at least not in time. The deviation surfaces at the release assay. The deviation report is written after the batch is gone. The post-mortem reviews the same historian data that was running live during the run — data that contained the early signal, if anyone had been watching for it.
That gap — between the data that was there and the detection that wasn’t — is the problem Praescia is built to close.
Why now
Three things changed in the last several years that make this the right moment:
First, the data is already there. Most manufacturing sites run process historians — OSIsoft PI and its successors, Honeywell PHD, and similar platforms. The signal that predicts a deviation is already being captured. The gap is interpretation, not instrumentation.
Second, the regulatory environment is pulling, not just permitting. FDA has actively encouraged PAT, continuous manufacturing, and real-time release for over a decade. The regulatory tailwind is real. A tool that improves batch-record quality and surfaces deviations earlier is aligned with where the FDA wants the industry to go.
Third, the composition pattern is proven. Praescia is built on Atlas — a composable platform that powers state-aware operating-spine products across manufacturing, industrial operations, legal, and buildings. The seven-stage spine (OBSERVE → NORMALIZE → DETECT_STATE → DETECT_TRANSITION → ESTIMATE_OUTCOME → RECOMMEND → VALIDATE) is a proven pattern. We are applying it to biopharma manufacturing, not inventing it for biopharma manufacturing.
What we built
Praescia composes two proven Atlas engines — Atlas PC (production constraint intelligence) and Atlas OI (industrial operations intelligence) — through the Atlas composable coupler, and wraps them in a GxP compliance shell.
The compliance shell is the one thing the sibling Atlas verticals don’t carry. Biopharma manufacturing is a regulated environment; any software touching a GMP line must honor 21 CFR Part 11, EU Annex 11, GAMP 5, and ALCOA+. So we built that in: an audit trail that is append-only and hash-chained, an e-signature framework that carries the signer, the time, the meaning, and a hash-link to the record, and a data-integrity gate at every I/O boundary that refuses malformed or unattributable records — fail-closed.
We call this arriving validation-ready. The tool ships with the compliance controls built in. We lower the customer’s computer-system validation burden; we don’t add to it.
What it does, concretely
The beachhead is upstream bioprocess: a live bioreactor run, monitored against a learned golden trajectory. Every run is residualized against the golden batch to make the non-stationary bioreactor signals tractable. The state machine tracks culture phase and golden-trajectory alignment. When the deviation onset is detected — the early metabolic shift, the DO deflection, the feed-rate divergence — the system surfaces an advisory: named signal, named reason, hours before the outcome is obvious.
The outcome estimate — predicted titer, yield band, OOS-risk probability — is called pre-harvest, not post-assay. That is the CFO-legible value: a saved batch is $100k–millions; a patient-specific cell therapy batch is one patient’s dose. The value of an earlier call is quantifiable.
What we’re honest about
The plumbing is built and tested. 73 tests pass on a synthetic bioprocess rig — a non-stationary bioreactor simulation with injectable labeled deviations. The compliance shell is implemented and reviewable. The validation package skeleton ships with the code.
But: synthetic data validates the plumbing. It does not prove detection. The evidence gate — the moment Praescia proves early deviation detection on real batch data from a real bioreactor — is what comes next. Until that gate is cleared, we make no detection claims beyond the architecture.
This is the regulated industry. Conservative, long-cycle, adversarial to overclaims. We are not going to win it by overstating where we are. We are going to win it by proving the detection on a real line and earning the reference.
What comes next: the design-partner program
We are looking for one CDMO or biotech MSAT team with real batch data to run a fixed-fee pilot on one process or line. You bring the data; we bring the foreknowing. If detection is real, we prove it together and build the reference. If it’s not, you get an honest answer and we improve the model.
That is the design-partner program. Apply here.
The name
Praescia (pray-SEE-sha) is coined from the Latin praescīre — to know beforehand — the root of prescience, foreknowledge. The foreknowing of the run. It sits naturally beside Atlas — where Atlas is the titan who bears the weight of the world, Praescia is the foresight that reads what is coming before it arrives, so the weight is never dropped.
We built it because a batch deserves foresight, not a post-mortem. Advising the team, never touching the equipment, and always on the record.
Praescia, powered by Atlas.