Sai Praneeth Kumar Molleti

Machine Learning and Data Science professional with 2 years of experience in data preprocessing, feature engineering, and model evaluation. Proficient in Python libraries such as NumPy, Pandas, and scikit-learn, with a focus on biomedical applications including EEG prediction and microscopy analysis. Skilled in developing production systems using AWS serverless technologies and multi-tenant backend solutions.

Title: Junior Machine Learning  Engineer

EEG Feature Engineering for Central Neuropathic Pain Prediction (SCI) | Python, scikit-learn, mlxtend

• Worked with high-dimensional EEG data (48 electrodes, 250 Hz; 432 features; N=18) to predict Central Neuropathic Pain development after spinal cord injury.
• Applied filter, wrapper, and embedded feature selection methods: Mutual Information, Fisher Score, L1/LASSO, and Forward Sequential Selection.
• Reduced feature space from 432 → ~170–175 while achieving mean cross-validated accuracy ≈ 0.88 using SVM (RBF) and Logistic Regression, with balanced sensitivity and specificity.
• Tuned models using GridSearchCV and evaluated trade-offs between accuracy, sensitivity, and specificity for a low-sample biomedical classification setting.

Model Selection for Clustering of Histopathology Patch Embeddings | Python, scikit-learn, PCA/UMAP

• Benchmarked K-Means, Gaussian Mixture Models, and Agglomerative Clustering on ~5,000 colorectal cancer tissue patch embeddings extracted from CNN representations.
• Evaluated clustering quality using Silhouette, V-measure, Davies–Bouldin, AMI, and Fowlkes–Mallows to avoid single-metric bias.
• Identified Agglomerative Clustering as the most consistent performer; PathologyGAN + PCA produced the clearest structure, while GMM underperformed and VGG16 + UMAP showed instability.

Fiber Segmentation & Morphology Analysis (M.Sc. Thesis — STORM Microscopy) | Python, scikit-image, SciPy, statsmodels

• Built an end-to-end microscopy image analysis pipeline to quantify fiber morphology across four biological conditions.
• Implemented segmentation using Canny edge detection, Otsu thresholding, and morphological post-processing.
• Extracted morphology features and validated group differences using one-way ANOVA with post-hoc tests; benchmarked classical ML models for interpretation.