Recursive AI for Chemical Formulation Discovery
A toolkit designed to accelerate pharmacological research through LLM-powered research agents, an advanced scoring function, and a formulation discovery algorithm.
The pharmaceutical and cosmetic industries invest significant time and resources in developing market-ready products. Researchers must analyze vast amounts of scientific literature, patents, and experimental data to develop effective and stable formulations.
Recursive AI for Chemical Formulations Discovery streamlines this process by generating innovative, regulation-compliant formulations ranked by key properties, significantly reducing development time and costs—all in just a few clicks.
1. Select Ingredients and Dosage Form: Start by choosing your active ingredient and specifying the quantity if needed. Then, select the preferred dosage form, such as a gel, cream, lotion, etc.
2. Customize Additives and Properties: Optionally add specific additives and fine-tune target physical and chemical properties based on your product requirements.
3. Generate Innovative Formulations: Our AI-driven discovery algorithm intelligently combines and mutates existing ingredients to produce novel formulations optimized for your target physical and chemical properties. This approach significantly reduces development time, eliminating weeks or even months of manual experimentation and iterative testing.
4. Receive Ranked Results: Instantly access a ranked list of generated formulations, complete with detailed scores for key properties and references to relevant research papers and patents. Providing structured and validated scientific insights empowers junior researchers with expert-level knowledge, enhancing their productivity and reducing reliance on senior guidance.
Highlights
- Extensive Ingredient Database: A foundation of over 2,900 known formulations and 2,400 ingredients provides a robust base for further discovery and optimization.
- LLM-based Research Agents: AI-powered agents mimic an expert pharmacologist conducting daily research, reading papers and patents, and extracting relevant information to further enhance the database.
- Improved Formulation Properties: Newly generated formulations are optimized by user-selected physical and chemical properties, such as those influencing product viability and onset of action.
- Regulatory Compliance: Ensures formulations meet pharmacopeia guidelines for dosage forms as defined by regulatory agencies.
- User-Friendly Web Platform: A clean, intuitive web interface gives researchers quick access to novel formulations with minimal effort.
Implementation

1. Data Source: Using FindFlow technology, Recursive has built a series of LLM agents that extract information from publicly available patents and research papers, integrating this information with an existing database of known formulations.
2. Data Preprocessing: The system encodes each ingredient using a chemical large language model, creating a high-dimensional representation that captures its chemical properties.
3. Bespoke AI Model:
- Scoring Function: Recursive has developed a deep-learning approach that integrates state-of-the-art AI models with bioinformatics tools to predict the physical and chemical properties of new formulations with 75% accuracy based on ingredient information and external conditions.
- Formulation Discovery Algorithm: The optimization agent generates new formulations by combining and mutating existing ingredients in novel ways to improve physical and chemical properties. Industry-specific knowledge is used to consider product requirements, constraints, and desirable dosage forms.
Related solutions
Typical implementation flow
Results within months
- 012 hours
Executive Briefing
- 022-4 weeks
Planning & Research
- 031-2 months
AI / Machine Learning Customization
- 04
Software Customization
Launch