Projects #

1. microfluidics-chemical-computation- June, 2026 #

Soft Hardware, Flowing Software: explainer, demo & research

A beginner-friendly explanation, a runnable software demo, and an in-progress computational research study built around the paper:

Soft Hardware, Flowing Software: Reconfigurable Microfluidics for Adaptable Chemical Computation Swinkels, Dúzs, Skarsetz, Nishiyama & Walther, Advanced Materials, June 2026. Advanced Materials (Wiley) · Open access (CC-BY)

2. LabVerify: Lab interface for Biotech Researchers operating OpenTron machines- June, 2026 #

A verification layer for agentic lab control. A scientist describes a liquid-handling protocol in plain English: labverify generates it, verifies it is mechanically correct before a real robot runs it, and drives the instrument. The headline check: it catches the silent error that ruins an assay (ask for 200 µL with a 50 µL tip, and the robot quietly splits it into 4×50 µL on the wrong tip) before anything moves.

3. Physics-Informed Neural Networks for Chemotherapy Pharmacokinetics: Benchmarking the Clinical Estimator and Exposing Parameter Identifiability- June, 2026 #

A small, self-contained PINN project that uses the 2-compartment pharmacokinetic model, the model that clinical pharmacologists actually use to dose doxorubicin, paclitaxel, methotrexate, cisplatin and similar cytotoxic chemotherapies and benchmarks a PINN against the standard clinical estimator and a physics-agnostic neural baseline.

4. Physics-Aware Auxiliary Losses Improve Out-of-Distribution Generalization of a GNN Synthesizability Filter- May, 2026 #

A graph neural network for predicting whether a molecule is synthesizable, with optional physics-grounded auxiliary losses (Bertz topological complexity + MMFF94 strain energy) layered on the GNN backbone. The research question: do physics-aware aux losses improve out-of-distribution generalization over a pure data-driven baseline?

5. The Metric Picks the Winner: Evaluation Choice Flips Model Rankings for Drug-Response Prediction in Unseen Chemistry- May, 2026 #

VCPI/Ginkgo Build-a-Virtual-Cell Hackathon 2026- Retrieval augmented chemistry-foundation models for drug-perturbation transcriptomes.

6. BinaryBrain-fork(Binary Neural Net framework)- May, 2025 #

Contributed to open source BinaryBrain framework- C++ framework based on LUTNet architecture for real time fast edge Inference on FPGAs. A training infrastructure for Binary Neural Networks. Github

7- Difflogic- A library for Differentiable Logic Gate Networks- March, 2025 #

Contributed to open-source logic gate based neural network library- difflogic(Based on the paper "Convolutional Differentiable Logic Gate Networks"- https://arxiv.org/abs/2411.04732. Github

8- Hardcoded Dataflow style Inference of Logic Gate Networks on FPGA(Part of Bachelors Thesis)- Jan, 2025 #

Designed a low-bit deep neural network Inference engine pipeline that uses UART-based RX/TX to send MNIST image inputs(28x28 and 20x20) to Efinix based FPGA and receive predictions within milliseconds via trained hardcoded Logic Gate Networks [Implementation of the paper "Differentiable Logic Gate Networks"(NeurIPS 2024)] Github

9- Machine Learning for Fundamental Physics- August, 2024 #

This repository contains all the notebooks used for the Machine Learning for Fundamental Physics (ML4FP) School 2024 at Lawrence Berkeley National Laboratory(LBNL). Topics covered are higgs classification, generative AI, introduction to binary classifier, anomaly detection, unfolding, etc. Github

10- Accelerating Deep Neural Networks(DNNs) Papers on FPGA #

Curated list of Deep Neural Networks on FPGAs research papers, covers Binary Neural Nets(BNNs), Logic Gate Networks(LGNs), and Look-Up Table Networks(LUTNets). Github

11- Exoplanet Atmospheric Characteristics- July, 2025 #

Researching exoplanets by applying Unsupervised Machine Learning Techniques like K-Means Clustering. Working with publicly available protoplanetary disk dataset by ALMA Observatory.