Media & seed train
Autonomous media and buffer prep and seed-train control for composition, sterility, and inoculum readiness.
Upstream control · cell culture & fermentation
Hold the bioreactor in its productive window.
Autonomous cell culture that reads the culture, predicts the drift, and moves feed and gas ahead of it, batch after batch.
Runs on Jetson Orin/Thor at the edge, DGX/HGX for training
The problem
Upstream culture is still nursed by hand.
A production bioreactor holds living CHO or microbial cells for days or weeks, and pH, dissolved oxygen, temperature, agitation, CO₂, and a timed feed all have to stay inside a narrow window. A small drift in feed or gas crashes viable cell density, starves the culture, or quietly shifts the glycosylation and charge variants that decide whether the drug is safe.
Today that culture is tuned by hand from setpoints and offline lab results, checked against golden-batch spreadsheets by a small and aging bench of process scientists. One crashed seed train or mistimed feed destroys a batch worth millions and idles a suite for weeks.
What Culteon does
Closed-loop control from media prep to the production reactor.
Culteon runs media and buffer prep and the seed train for composition, sterility, and inoculum readiness, then takes control of the production reactor. It reads the live culture, predicts where cell density and metabolites are heading, and adjusts feed and gas before the drift instead of chasing it. Every action is bounded, supervised where the process requires it, and written to a validated audit trail.
Production control
Closed loop on pH, DO, temperature, agitation, CO₂, and fed-batch or perfusion feed at peak viable cell density.
Predictive feed
Acts ahead of cell-density and metabolite drift rather than reacting to a lab result hours later.
Graduated autonomy
Human-in-the-loop checkpoints and a full 21 CFR Part 11 audit trail on every decision.
Fits your stack
Connectors to bioreactor DCS and OPC, single-use skids, and MES such as PAS-X and Syncade.
Learns across trains
Control learned on one train sharpens control on every train through the shared data network.
- Time-series prognostics for VCD, glucose, lactate, ammonia, and viability
- Hybrid mechanistic and ML kinetic models behind the feed and perfusion policy
- Reinforcement and optimization-based feed and gas control tuned per cell line
- SOP-grounded reasoning for batch context and deviations
Under the hood
The AI that steers the culture.
Setpoint controllers hold a number. Soft-sensor tools predict but never touch the feed. Culteon closes the perceive-decide-act loop on the production reactor and gets better with every batch it runs. Per-cell-line control policies improve as they train on partner data.
Built on NVIDIA
GPU-essential at the edge and in training.
- Edge hardwareJetson Orin and Thor (IGX) run the low-latency control loop next to the bioreactor, where a feed action cannot wait for a remote call.
- TrainingDGX and HGX train and fine-tune per-cell-line kinetic and control models over batch histories.
- LibrariesCUDA and TensorRT accelerate on-edge inference. BioNeMo grounds the cell-line and metabolite modeling. cuOpt optimizes feed and perfusion windows under equipment and quality constraints.
- ServingTriton and NIM serve the models on site GPUs under NVIDIA AI Enterprise for auditable, vendor-supported GMP inference.
Commercials
Priced per train, expands per site.
- PricingBioseon Line module at $13,000 per bioreactor per month, expanding into the Site tier. Usage metering per batch above plan, and outcome pricing on titer and yield lift for enterprise.
- Who buysBiomanufacturing and MSAT directors hold the budget, upstream process engineers champion it, QA and PAT managers approve.
- The moatA compounding cross-site dataset of bioreactor responses and feed and perfusion trajectories that no single-task controller can assemble.
- First deploymentRead-only monitoring and prediction on one train, then advisory feed recommendations, then bounded closed-loop control under human approval.
Where it fits
Culteon owns upstream. It does not work alone.
Put one bioreactor on Culteon.
We connect to a single train, calibrate against your historical batches, and run in shadow mode before closing the loop. Most lines reach closed-loop control within a validation cycle.