LLM-driven rapid prototyping of microfluidic devices: a deterministic Python-mediated framework for design, fabrication, and validation
Muhammad Aneeq Nazim Khatana, Muhammad Aqib Raza Shah, Takayuki Shibata, Shunya Okamoto, Tuhin Subhra Santra, Moeto NagaiPurpose
Microfluidic device design remains time-intensive and heavily reliant on computer-aided design (CAD) expertise, which limits scalability in rapid prototyping, high-throughput experimentation and standardized design workflows. While large language models (LLMs) offer a route toward automated CAD, existing script-based CAD tools, e.g. CADQuery, are prompt-sensitive and prone to geometric inconsistencies. This study aims to enable reliable rapid prototyping by accelerating the design-to-fabrication cycle through an LLM-assisted, Python-mediated framework for microfluidics and performing physical fabrication and dimensional validation to confirm whether designs meet manufacturing tolerances and functional requirements.
Design/methodology/approach
ChatGPT is used to generate a closed-loop Python script, executed through a Python script that generates scalable vector graphics (SVG), drawing exchange format (DXF) and standard tessellation language (STL) outputs of microfluidic geometries, including Spiral Mixers, T-Junctions and Y-Junctions along with script execution time. Each design run is logged with a cryptographic hash to ensure traceability and confirm reproducibility. Physical fabrication using fused deposition modelling and dimensional validation using optical metrology on a digital microscope were essential to confirm whether designs meet manufacturing tolerances and functional requirements. Device functionality was assessed through pressure resistance evaluation, with indicative pressure-drop estimates calculated at leakage onset using the Hagen–Poiseuille approximation.
Findings
ChatGPT/Python workflow substantially reduces iteration time compared with the manual CAD workflow. Post-fabrication measurements showed dimensional deviations up to 12.6% across all geometries while iterative compensation further reduced dimensional deviations. Pressure resistance testing confirmed indicative functional device integrity within low-to-moderate flow regimes, with leakage onset observed at flow rates consistent with fused deposition modelling-related sealing limitations, with no evidence of gross geometric errors in the generated CAD files.
Originality/value
This workflow separates probabilistic LLM interaction from deterministic geometry execution. It provides a traceable route from design generation to exploratory fabrication, supported by cryptographic logging, dimensional validation and iterative compensation.