Dimitrios Tsaoulidis
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Dimitrios Tsaoulidis has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
Tao Chen
3 shared publications- Tao Chen
School of Chemistry and Chemical Engineering, University of Surrey, Guildford GU2 7XH, UK
3 shared publications - Yongrui Xiao
School of Chemistry and Chemical Engineering, University of Surrey, Guildford GU2 7XH, UK
3 shared publications - Yu Zhang
School of Chemistry and Chemical Engineering, University of Surrey, Guildford GU2 7XH, UK
3 shared publications Anna Tsitouridou
2 shared publicationsChuan-Yu Wu
2 shared publications
Latest Publications
- Open Access
- 18 October 2026
- Published by Elsevier in SSRN Electronic Journal
SkinMiner: LLM-powered data mining to accelerate formulation design for skin products
Evaluating dermal and transdermal formulations requires costly and time-consuming permeation experiments, making the reuse of existing experimental data highly valuable. However, leveraging these data remains challenging because relevant measurements are scattered across the text, tables, and figures of decades of published studies. Motivated by the strong reasoning and information-understanding capabilities of large language models (LLMs), we propose SkinMiner that leverages LLMs to extract structured and verified multimodal permeation records from relevant literature. SkinMiner introduces a modular architecture that reads each study across text, table, and figure modalities. A single user-supplied Target Specification controls scope consistently at both retrieval and acceptance, and an evidence-grounded verification stage admits a record only when it is in scope and linked to located source evidence, giving every value a provenance trail. On an ibuprofen evaluation, the verified output contained only records that fell within the requested scope, and 96% of extracted values matched the source, judged against a manually annotated gold standard. Compared with a single unconstrained model call, SkinMiner gave consistent results across two LLMs and produced far less unverified output. Two demonstrations then used the extracted data in formulation design workflows: records from a few papers ranked and approximately predicted fluxes in a related cosolvent system, and a mixing-ratio sweep quantified, before any model was fit, whether literature records would reduce the in-house experiments needed. SkinMiner turns published permeation data into verified, provenance-linked records that can be assembled at literature-scale without manual transcription, making prior experimental evidence directly usable in formulation design.
- 18 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSkinMiner: LLM-powered data mining to accelerate formulation design for skin products
Evaluating dermal and transdermal formulations requires costly and time-consuming permeation experiments, making the reuse of existing experimental data highly valuable. However, leveraging these data remains challenging because relevant measurements are scattered across the text, tables, and figures of decades of published studies. Motivated by the strong reasoning and information-understanding capabilities of large language models (LLMs), we propose SkinMiner that leverages LLMs to extract structured and verified multimodal permeation records from relevant literature. SkinMiner introduces a modular architecture that reads each study across text, table, and figure modalities. A single user-supplied Target Specification controls scope consistently at both retrieval and acceptance, and an evidence-grounded verification stage admits a record only when it is in scope and linked to located source evidence, giving every value a provenance trail. On an ibuprofen evaluation, the verified output contained only records that fell within the requested scope, and 96% of extracted values matched the source, judged against a manually annotated gold standard. Compared with a single unconstrained model call, SkinMiner gave consistent results across two LLMs and produced far less unverified output. Two demonstrations then used the extracted data in formulation design workflows: records from a few papers ranked and approximately predicted fluxes in a related cosolvent system, and a mixing-ratio sweep quantified, before any model was fit, whether literature records would reduce the in-house experiments needed. SkinMiner turns published permeation data into verified, provenance-linked records that can be assembled at literature-scale without manual transcription, making prior experimental evidence directly usable in formulation design.
- 15 July 2026
- Published by Elsevier in Chemical Engineering Research and Design
- Vol. 232, 403-416
Active learning-based adaptive optimisation for developing dermal drug formulations
- 15 July 2026
- Published by Elsevier in Chemical Engineering Research and Design
- Vol. 232, 403-416
Active learning-based adaptive optimisation for developing dermal drug formulations
- Open Access
- 9 June 2026
- Published by Royal Society of Chemistry (RSC) in RSC Pharmaceutics
- Vol. 3 (4) , 1034-1043
A modelling-based comparison of IVRT and synthetic-membrane permeation for early formulation screening
IVRT release rates (6 h, nylon membrane) correlate strongly with IVPT flux (30 h, Strat-M) across fifteen ibuprofen poloxamer-407 formulations (Pearson r = 0.95, R 2 = 0.92), supporting IVRT for early formulation screening.
- 9 June 2026
- Published by Royal Society of Chemistry (RSC) in RSC Pharmaceutics
- Vol. 3 (4) , 1034-1043
Open AccessA modelling-based comparison of IVRT and synthetic-membrane permeation for early formulation screening
IVRT release rates (6 h, nylon membrane) correlate strongly with IVPT flux (30 h, Strat-M) across fifteen ibuprofen poloxamer-407 formulations (Pearson r = 0.95, R 2 = 0.92), supporting IVRT for early formulation screening.
- 10 May 2026
- Published by Elsevier in Chemical Engineering Research and Design
- Vol. 230, 756-766
Confidence-bound early stopping of experiments with sequential calibration
- 10 May 2026
- Published by Elsevier in Chemical Engineering Research and Design
- Vol. 230, 756-766
Confidence-bound early stopping of experiments with sequential calibration
- Open Access
- 18 February 2026
- Published by Elsevier in SSRN Electronic Journal
Controlled PLGA microparticle production across different microfluidic channel sizes
Precise control of microparticle synthesis is essential for advanced drug delivery systems. Microfluidics offers improved controllability and particle quality, yet challenges remain in scalability and the consistent production of small (< 5μm) particles without extensive post-processing. This study develops a generalized framework for controlling microparticle size and uniformity across different microfluidic device scales. Poly (lactic-co-glycolic acid) (PLGA) microparticles were produced using two geometrically similar flow-focusing microfluidic devices (30 μm and 100 μm). By directly comparing droplet formation dynamics and particle characteristics, we derive transferable insights and dimensionless design rules. Key operational parameters (i.e., surfactant concentration, polymer concentration, flow rate ratio (FRR), fluid velocity, and residence time) were varied to assess their impact on droplet and particle characteristics (i.e., size and polydispersity). Results revealed that increasing surfactant concentration and FRR yields smaller, more monodisperse particles, whereas increasing polymer concentration leads to larger particles and higher polydispersity. The effect of FRR in the 30 μm device was ~45% greater, producing smaller (2–11 μm) and more uniform particles compared to those in the 100 μm device (9–25 μm). A predictive model for particle size based on Capillary number, Reynolds number, and dimensionless droplet size achieved an average deviation of 6.2%.
- 18 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessControlled PLGA microparticle production across different microfluidic channel sizes
Precise control of microparticle synthesis is essential for advanced drug delivery systems. Microfluidics offers improved controllability and particle quality, yet challenges remain in scalability and the consistent production of small (< 5μm) particles without extensive post-processing. This study develops a generalized framework for controlling microparticle size and uniformity across different microfluidic device scales. Poly (lactic-co-glycolic acid) (PLGA) microparticles were produced using two geometrically similar flow-focusing microfluidic devices (30 μm and 100 μm). By directly comparing droplet formation dynamics and particle characteristics, we derive transferable insights and dimensionless design rules. Key operational parameters (i.e., surfactant concentration, polymer concentration, flow rate ratio (FRR), fluid velocity, and residence time) were varied to assess their impact on droplet and particle characteristics (i.e., size and polydispersity). Results revealed that increasing surfactant concentration and FRR yields smaller, more monodisperse particles, whereas increasing polymer concentration leads to larger particles and higher polydispersity. The effect of FRR in the 30 μm device was ~45% greater, producing smaller (2–11 μm) and more uniform particles compared to those in the 100 μm device (9–25 μm). A predictive model for particle size based on Capillary number, Reynolds number, and dimensionless droplet size achieved an average deviation of 6.2%.
- Open Access
- 1 September 2025
- Published by Elsevier in Chemical Engineering Journal Advances
- Vol. 24
CFD approaches in microfluidics for the development of polymeric, lipid-based and inorganic nanoparticles for drug delivery
- 1 September 2025
- Published by Elsevier in Chemical Engineering Journal Advances
- Vol. 24
Open AccessCFD approaches in microfluidics for the development of polymeric, lipid-based and inorganic nanoparticles for drug delivery