Ahtesham Alvi

IBD Miniprotein Binder Design Pipeline

University of Maryland, Department of Computer Science

Undergraduate Researcher · March 2026 — Present

Python

Developing a computational pipeline for designing miniprotein binders (5–15 kDa) targeting IL-23 and alpha-4-beta-7 integrin for inflammatory bowel disease (IBD) therapeutics.

• Implemented a 9-stage structure-based protein design pipeline integrating RFdiffusion for de novo backbone generation, ProteinMPNN for inverse folding sequence design, and AlphaFold2-Multimer for structural validation.
• Built hotspot identification and pocket extraction modules to map binding epitopes and encode geometric constraints for diffusion-based protein generation.
• Developed backbone filtering and binding evaluation stages scoring interface quality via buried surface area, shape complementarity, and binding free energy (ddG).
• Engineered multi-objective candidate ranking with diversity filtering to select optimal therapeutic binder candidates.

Exoplanet Habitability Classification

University of Maryland, College Park

Independent Research · January 2026 — Present

Python

PyTorch

Scikit-learn

Developing neural and embedding-based ML models to predict exoplanet habitability using NASA and JPL datasets, projecting catalog entries into a learned habitability space for similarity-based classification.

• Engineering feature encodings of planetary parameters and training custom neural networks for similarity-based habitability classification.
• Identifying candidate habitable exoplanets by projecting NASA catalog entries into a learned habitability embedding space.
• Processing and cleaning multi-source astronomical datasets from NASA Exoplanet Archive and JPL for model training and validation.

Temporal Robustness of Social Media Bot Detection Models

University of Maryland, College Park

Undergraduate Researcher · February 2026 — Present

Python

PyTorch

Scikit-learn

Investigating the temporal generalization of social media bot detection models by evaluating whether classifiers trained on historical Twitter (X) datasets remain effective against modern bot behaviors.

• Designed a machine learning pipeline using the Twibot-22 dataset (1M+ users, 86M tweets) to train classifiers for bot vs. human detection based on behavioral, linguistic, and network features.
• Analyzed model generalization by applying trained classifiers to modern X data, quantifying performance degradation across temporal distribution shifts.
• Conducted feature importance and behavioral analysis to identify which detection signals remain stable versus those that lose predictive power over time.
• Evaluated models using accuracy, precision, recall, and F1-score, and investigated failure modes to distinguish overfitting from true evolution in bot strategies.
• Explored adaptive modeling strategies and feature redesign approaches to improve robustness of bot detection systems in evolving social environments.

Bioinspired Robotics — Flapping Winged Glider (RoboRaptor)

FIRE: First-Year Innovation & Research Experience, University of Maryland

Undergraduate Research Intern · May 2024 — July 2024

Arduino

C++

AutoCAD

Autodesk

Led development of a bio-inspired articulated flapping-wing robotic glider integrating servo actuation, lightweight aeroelastic airframe, onboard sensing, and telemetry for wind–flight interaction research.

• Engineered Arduino Nano firmware unifying flapping control, torque modulation, remote receiver interfacing, sensor acquisition, and ~30 Hz OpenLog SD-card telemetry logging.
• Designed and iterated carbon-fiber wing spars and 3D-printed servo mounts to withstand cyclic flapping loads, vibration, and torque stresses.
• Built wind and inertial sensing pipeline to record environmental and flight dynamics data for analysis of aeroelastic behavior.
• Conducted wind-dependent flight experiments validating stable tethered gliding, servo-driven flapping, and directional control across multiple prototype iterations.