This project explores the integration of generative AI into pharmaceutical crystallization control, tackling modern manufacturing challenges and enabling intelligent, autonomous process optimization.
Pharmaceutical Crystallization Challenges
Crystallization is essential in producing active pharmaceutical ingredients (APIs), determining purity, particle size, shape, and downstream performance. Yet, as detailed in a recent commentary from Nature Chemical Engineering A Changing Paradigm in Industrial Pharmaceutical Crystallization, the field is undergoing a shift:
- Active Pharmaceutical Ingredients (APIs) are becoming more complex—moving from small molecules to peptides and macromolecules.
- Crystallization scale-up is more difficult for these large, often amorphous compounds.
- Predicting solubility, morphology, or nucleation behavior remains a challenge.
- Traditional trial-and-error development cycles are inefficient for rapid drug development.
This creates an urgent need for smarter control strategies—ones that can adapt in real time, learn from data, and incorporate physical knowledge.
Mission Statement
To develop a next-generation model predictive control (MPC) framework empowered by generative AI for real-time, self-improving control of pharmaceutical crystallization processes.
Project Objectives and Methods
1. Literature Review
- Investigate the state-of-the-art in MPC for batch and continuous crystallization.
- Explore transformer-based models and physics-informed neural networks (PINNs) for process modeling.
2. Develop GenAI Agents
- Create AI agents that automatically build predictive models from process data.
- Use transformers to characterize hard-to-model properties like crystal size and shape distributions.
- Implement in a simulated environment (e.g., PharmaPy) for closed-loop AI control.
- Create a digital twin of the crystallization system.
- Use GenAI-driven MPC to meet defined quality and performance targets.
- Develop interpretable linear models for predictive control.
- Focus on explainability and transparency.
Phase II: Batch Crystallization
- Apply nonlinear and transformer-based models to predict critical quality attributes (CQAs).
3. Human-in-the-Loop Interaction
- Enable operators to input objective functions (e.g., desired crystal size or morphology).
- AI agents provide optimized control settings that fulfill those objectives.
4. Closed-Loop Demonstration
- Test AI-MPC performance in simulated environments with real-time feedback loops.
5. Integration with Control Platforms
- Embed GenAI models into existing MPC platforms.
- Enable real-time optimization of crystallization and drying operations.
6. Autonomous Learning and Optimization
- Use GenAI agents to adapt control strategies continuously based on live process data.
- Demonstrate gains in yield, consistency, and efficiency compared to conventional methods.
Undergraduate Research Opportunity
- Position: Undergraduate Research Assistant – Chemical Engineering / Machine Learning
- Hours/Week: 5–20
- Pay: Based on academic year (Freshman, Sophomore, Junior, Senior)
- Work Environment: On-campus and remote options available. Collaborative environment with regular mentorship.
Key Responsibilities:
- Assist in developing AI/ML models for pharmaceutical crystallization processes.
- Support data collection, model training, and simulation in environments like PharmaPy.
- Work with large datasets and advanced visualization tools.
Preferred Qualifications:
- Enrolled at BYU in Chemical Engineering, Data Science, or related field.
- Experience or interest in Python, AI/ML frameworks, or process systems modeling.
- Curiosity and drive to solve complex, real-world engineering problems.
To apply, contact Jonathan Pershing or visit the PRISM Group site for more info.
Graduate Student Lead: Jonathan Pershing
Jonathan Pershing is a PRISM Group researcher in Chemical Engineering. He leads the AI-Augmented Crystallization Project and specializes in model predictive control, process modeling, and generative AI. Jonathan is passionate about tackling the unthinkably difficult through engineering, persistence, and innovation.
Research Assistant: Megan Booth
Megan Booth is an undergraduate research assistant contributing to the AI-Augmented Crystallization Project through her expertise in crystallization techniques and process optimization. Her background includes previous research in terahertz spectroscopy and organic synthesis, as well as leadership and service experience in Guatemala. Passionate about applications engineering, automation, and process control, Megan is driven by the opportunity to apply emerging technologies to real-world industrial challenges.
Stay Connected
Visit this page for updates on progress, results, and opportunities to get involved. The project will share findings, simulations, and open-source model components to advance the field of AI-integrated process control.