Chemists' preferences for specific molecules are highly context-dependent, varying significantly based on the specific projects they are engaged in. In an organizational setting, diverse chemists interact with various projects, providing valuable feedback on their molecular preferences. This research explores the potential of leveraging meta-learning to minimize the required feedback for accurate preference learning. By treating the task of understanding a new chemist's or new project's molecular preferences as a few-shot learning problem, we can significantly reduce the feedback burden. We propose the use of (latent) neural processes as a solution, enabling efficient and rapid adaptation to new chemists' preferences or project contexts with minimal feedback. Here, existing feedback datasets are mapped into a latent space, using a VAE setup. The predictions of preference are then made conditional on the latent space vector. This approach promises to enhance the efficiency and accuracy of preference learning in dynamic and diverse chemical research environments. In the initial experiments, we simulate chemists interacting with different project to generate the simulated feedback data.