Building an AI-Powered Formulation Platform for the Chemical Industry

Name: Emerline
Brand: Emerline
Rating: 4.9 (25 reviews)

Emerline developed a custom AI tool that enables companies to train models on their own formulation test results, speeding up R&D processes in the chemical and cosmetics industries.

FastFormulator is a company that provides an R&D platform to help businesses in the chemical, food, and cosmetic industries improve product design. It uses AI and machine learning to suggest component combinations and ratios that match specific physical or physiochemical targets, such as viscosity or surface tension, thus optimizing product formulations.

Their mission is to reduce the time and effort chemists and other scientists spend on manual experiments. In traditional workflows, researchers often need to run dozens of iterations, adjusting components and proportions until they reach the desired product characteristics — a process that’s time-consuming and costly.

AI-based tools can help reduce the number of experiment iterations, saving time and money. However, most available solutions are not built for formulation tasks and often produce unreliable results. FastFormulator saw this gap and set out to develop a more accurate, domain-aware system.

Our approach

We set out to build a web-based platform that could help researchers accelerate formulation design with the help of machine learning. To achieve this, we focused on several practical goals from the start.

Our role as the development team was to:

Build a web application where users could upload their experimental results.

Connect the app with a machine learning engine to train models based on each customer’s data.

Enable users to receive accurate, data-driven suggestions for new formulations.

Make it possible to customize component lists and targets, industry and use cases.

Ensure secure data separation between clients.

Implement user access and permission control.

Also, throughout development, we had to address the typical R&D pain points:

High costs & time consumption: Replacing dozens of manual experiments with data-based predictions.

Unpredictable outcomes: Providing reliable suggestions instead of trial-and-error.

Scattered data: Structuring customer-provided data through a clean and consistent interface.

With the idea for a specialized platform in place, the client outlined the main challenges they needed to address:

Too many manual experiments
High R&D costs
Long lead times to reach target product properties

The client needed a flexible, industry-aware tool that could adapt to various formulation scenarios — from food to cosmetics — without compromising prediction quality or user experience.

To make that possible, the platform had to:

Accept experimental data in a structured, accessible way.
Learn from each client’s input and adapt predictions accordingly.
Keep data isolated and secure for each customer.
Provide a clean, easy-to-use interface for users without a data science background.

It also had to manage out-of-the-box components, like surfactants or oils, using an admin interface, while letting each customer upload their own substances with chemical formulas using SMILES notation (in a computer-readable format).

These requirements defined the technical direction for the solution, covering overall architectural choices, development tools and frameworks, and plans for the subsequent development process.

Solution

Each company can train its own models within the platform using internal data, without risking exposure. This ensures both data security and result accuracy, as predictions are based on each customer’s real-world experiments rather than generalized datasets.

The solution includes several key features:

This feature allows companies to manage customers, including their available model finetunings and user invitations, and edit shared component dictionaries.

Includes account invitation, password reset, and login/logout flows.

Each company can manage their own team, assigning permissions and access levels.

Users can upload experimental results (e.g., components and percentages vs. output properties) to train models specific to their use case. Multiple model versions can be created for each dataset.

Clients can upload their own substances along with chemical formulas in SMILES format to use in formulations.

A simulation space where users select components and their ratios to see predicted formulation properties. It helps fine-tune inputs to match target characteristics.

This tab automatically generates formulations optimized for chosen targets, showing the most promising combinations and allowing users to refine them further.

Users can define and adjust their list of ingredients or targets, such as foaming, stability, surface tension, or viscosity.

The platform uses machine learning to identify the best-performing formulation — including both the choice of components and their ratio — that is most likely to reach the desired properties.

Compared to traditional models, FastFormulator’s approach reduces the amount of training data needed.

The ML engine is deployed as a containerized service and remains under the control of each customer, meaning the client can develop and fine-tune their models securely and independently.

Technology Stack

Back-end (Main Service)

Python

Django

Celery

PostgreSQL

AWS S3

Front-end

React

Redux

Machine learning services

Python

FastAPI

PyTorch

PyTorch Lightning

Scikit-learn

We successfully delivered a digital platform that helps chemists move from manual experimentation to faster, data-driven formulation.

Our solution enabled faster, cost-wise R&D and accurate outcomes, as follows:

Provided chemists with relevant formulation suggestions based on the real experimental data.
Reduced the need for repeated manual tests, saving time and resources.
Helped users quickly evaluate different component combinations without running physical experiments.
Afforded a clean, secure, and accessible environment for working with proprietary data.
Ensured that all data remained customer-specific: predictions from one company’s model were not influenced by data from others.

With FastFormulator, users can approach product development with more confidence and run their experiments more efficiently. The platform sets the stage for smarter R&D workflows while remaining flexible enough to support use cases across multiple industries.

Customer review

We are proud to have earned a reputation of a reliable software development provider, which is supported by the feedback received from our client.

More Case Studies

AI-powered Document Search Service for B2B Customers

Emerline has developed an innovative AI-powered chatbot document search service tailored for B2B customers, designed to significantly enhance document management and retrieval processes in a business environment.

Data Science

Machine Learning

Custom development Software & Hi-tech AI ML Chatbots