RafflesiaRequest access

01 Platform

Intelligent environments that automate discovery.

Rafflesia gives AI labs Gymnasium-compatible simulated wet lab environments — pipetting, cell culture, plate handling — so RL agents can learn experimental design without consuming real reagents or equipment time.

Liquid HandlingPlate TransferCell CultureImagingGymnasium-compatible

6

environments

at launch

<5ms

step latency

per simulation step

10×

faster iteration

vs. real wet lab

100%

Gym-compatible

drop-in for any RL loop

02 Core capabilities

Built to train agents on real biology.

High-fidelity simulation

Environments grounded in real wet lab physics and stochastic noise profiles. Train agents on the same variability they will face in real experiments.

Gymnasium-compatible API

Drop-in compatibility with standard RL frameworks. Structured action spaces, dense reward signals, and reproducible seeds for fair policy comparison.

Standardised benchmarks

Published baselines and leaderboards across modalities. Measure how well a policy generalises across plate formats, reagent concentrations, and noise levels.

03 Environments

Modalities available today.

Liquid Handling v1

Pipetting
Available

Success rate

0.94

Plate Transfer v2

Plate Handling
Available

Reward (↑)

1.21

Cell Culture v1

Culturing
Beta

Success rate

0.87

Widefield Imaging v1

Imaging
Beta

Success rate

0.91

Synthesis v1

Chemistry
Coming soon

Reward (↑)

Bioprocess v1

Fermentation
Coming soon

Success rate

04 How it works

From framework to first rollout in minutes.

01

Connect your framework

Install the Rafflesia client. Works with Stable Baselines3, RLlib, CleanRL, and any framework that speaks the Gymnasium interface.

pip install rafflesia
import rafflesia as gym
02

Pick an environment

Choose a lab modality and configure your plate format, reagent concentrations, and noise profile.

env = gym.make("LiquidHandling-v1",
  plate="96-well",
  noise=0.02)
03

Train and benchmark

Run your policy loop. Metrics stream to the dashboard. Compare against published baselines or share runs with collaborators.

obs, _ = env.reset()
for step in range(steps):
  action = policy(obs)
  obs, r, done, _ = env.step(action)

05 Who it's for

For labs that train agents, not pipettes.

Frontier AI labs

Build RL agents that can plan and execute multi-step biological experiments. Replace costly wet lab iteration with fast simulation rollouts.

AI-bio startups

Accelerate drug discovery, bioprocess optimisation, and automated synthesis with agents trained entirely in simulation before touching real equipment.

Research teams

Reproduce results, share environments, and compare policies with standardised benchmarks — the missing infrastructure layer for AI biology research.

06 Early access

We're working with a small group of founding labs.

If your team is building RL agents for biological applications and needs high-fidelity simulation environments, we want to hear from you. Early access partners shape the roadmap.