Hi, I'm Farzana!

Persona-MCMC: Iterative LLM Persona Refinement for Interpretable Student Modeling

What if DLA has it backwards? Flow reverses aggregation from divergent to convergent growth

Protein–Ligand Affinity Score Using Motion Planning

Development of tools for parameterization and validation of a model of fractal growth using image analysis and embedding

• Working on a geospatial traffic-data pipeline that retrieves Annual Average Daily Traffic (AADT) from transportation data sources using geographic coordinates, GIS APIs, and spatial roadway matching.
• Implementing roadway and intersection matching using Python-based geospatial processing and roadway geometry to associate user-selected locations with corresponding traffic data and roadway segments.
• Investigating computer vision and OCR methods for extracting and validating traffic information from transportation maps, with emphasis on improving reliability for integration into production transportation software.

• Developing a machine learning approach for estimating intersection turning movement counts (TMCs) from available traffic and roadway data, including data processing, feature development, model evaluation, and analysis of estimation performance.

• Designed a Metropolis-Hastings-inspired Markov chain Monte Carlo (MCMC) pipeline to generate interpretable student personas and predict students' performance.
• Worked on prompt engineering, built supporting baselines and an iteration logging system to compare against In-Context Learning (ICL) and Item Response Theory (IRT) models.

• Worked on predicting protein–ligand binding affinity scores using a modified Probabilistic Roadmap (PRM) motion planning algorithm, performed energy-based evaluation using the OpenMM molecular dynamics library, and benchmarked results against AutoDock and CAVER to analyze geometric and energetic constraints.

• Investigated the behavior of a triangular agent-based model to simulate fractal growth similar to absorptive spin-coating polymers.
• Generated statistics using ImageJ/Fiji with Mitochondrial Network Analysis (MiNA) preprocessing and developed an automated algorithm to calculate angles between polymer branches.

• Tutored and graded coursework for two undergraduate courses: Introduction to Statistics and Linear Algebra.
• Assisted peers through homework support and feedback, and helped develop problem-solving skills.

• Performed data analysis on 300k+ rows of data to identify taxi market trends, and collaborated on data preprocessing using several cleaning techniques.
• Built and evaluated ML models (Logistic Regression, Random Forest) for medical condition prediction, deploying the final model as a Flask web app on Heroku.

• Analyzed sales data to identify factors driving successful leads using Random Forest, k-means clustering, and Tableau.

This project asked two questions: can a fine-tuned BERT model tell whether a student's answer to a science question is correct, and do its attention weights point to the words that matter? I fine-tuned BERT-base on SciEntsBank (about 11K student responses to 197 science questions across 15 domains) for correct/incorrect classification. I evaluated it on three increasingly hard test sets: unseen answers, unseen questions, and unseen domains. I then visualized the [CLS] token's attention for correct and incorrect answers. Class imbalance pushed the model toward predicting "incorrect," and attention focused on content words in the question without clearly separating right from wrong answers. This result motivated my interest in looking deeper inside transformers. Next steps include adding the reference answer as input and trying 3-way labels.

Tools: Python, BERT-base (fine-tuning), SciEntsBank, attention visualization · View slides →

Built an AI platform using GPT-4 and RAG to prioritize appointments by urgency, helping counselors manage caseloads efficiently. Integrated with a calendar API to automate scheduling and notifications.

Tools: Python, GPT-4, RAG, Flask, HTML, Tailwind CSS, API integration · GitHub →

Developed a web app providing real-time forex market insights through a dynamic, interactive dashboard. Integrated live data from yFinance and the OANDA API, with predictive analytics to support informed trading decisions.

Tools: Python, Flask backend, Vue.js for a responsive UI · Devpost →

Analyzed 3,000+ behavioral records to study how smartphone usage (unlock patterns, night activity, and momentary assessments) relates to attention across a semester. Built a mixed-effects regression pipeline and implemented per-user ARIMA and SARIMA time-series models to capture temporal structure in productivity patterns. Evaluated the significance of predictors (p-values, F-statistics) and computed per-user RMSE for model comparison.

Tools: Python, pandas, statsmodels, scikit-learn, data visualization · GitHub →

CNN classifying cat poses, using image preprocessing and data augmentation (66% validation accuracy).

Tools: TensorFlow, Keras · GitHub →

Awarded to a student who shows exceptional perseverance, resilience, and dedication to academic and community excellence.

Awarded for academic excellence and commitment to advancing research. It supported full-time summer research in mathematical modeling.

Selected for a competitive fellowship on leadership development, negotiation, and strategic communication. Completed intensive workshops with industry leaders and worked with peers on case-based problem solving.

Recognized for outstanding contributions to campus community building, cross-cultural engagement, and mentorship through student organizations and residence life.

Led cultural programming, community engagement, and collaboration with the Bondhu Organization to promote South Asian representation on campus. Organized large-scale cultural events, fundraisers, and workshops, increasing membership and visibility.

Managed a residential community of 48 students, fostering an inclusive and supportive environment. Organized themed programming, mediated conflicts, enforced policy, and supported student well-being through peer mentorship and crisis-response training.

Designed promotional materials, managed outreach campaigns, and increased engagement for CS events, workshops, and study sessions. Developed branding and digital content to support community growth and visibility.

Represented student voices in institutional decision-making. Advocated for academic and student-life improvements, contributed to policy discussions, and collaborated with committees to strengthen campus initiatives.