Distributed intelligence
Scalable pipelines and data architectures that support machine learning without sacrificing reproducibility or operational clarity.
RESEARCH / 04
Questions that make
engineering more trustworthy.
My research explores how distributed and intelligent systems can remain reliable, interpretable and useful in high-stakes environments.
Trustworthy monitoring
Drift, anomaly and fairness-aware methods that detect when data or models stop behaving as expected.
Healthcare analytics
Privacy-conscious systems for clinical data integrity, medical imaging and decision support.
RESEARCH PREPRINT / 2025
Generation of Automatic Image Captioning Using Natural Language Processing
Deepala, V.K., Buttul, S.F., & Pally, C.
IEEE CONFERENCE PAPER / 2020
Pandemic Stabilizer using Smartwatch
Deepala, V.K., et al. · Contributed across research, system design, technical writing and peer mentoring.
CURRENT RESEARCH / MANUSCRIPTS
Narrative-Guided Interpretable Deep Learning for Early PTSD Prognosis
Developed the NG-IDL framework to combine structured trauma data with clinician-readable narratives through semantic mapping, cross-modal attention and interpretable prediction.
- 20,000 samples
- NG-IDL + SNSM
- AUROC · AUPRC · F1
Uncertainty-Aware Multimodal Fusion for Robust Glioma Grade Classification
Built a lightweight four-branch EfficientNet-B3 model with attention fusion and MC Dropout to classify LGG versus HGG when MRI protocols are incomplete.
- BraTS 2021 · 1,251 cases
- AUC 0.8919 ± 0.0167
- 15 missing-modality tests
- 1.94× uncertainty signal
ACADEMIC WORK / SELECTED STUDIES
Breast Cancer Prediction
Exploratory analysis and supervised classification of diagnostic features, comparing Logistic Regression, Random Forest, Decision Tree and SVM with cross-validation and hyperparameter tuning.
Open research repositoryCliff Walking: SARSA vs Q-Learning
A grid-world study comparing the safer on-policy route learned by SARSA with the optimal but risk-exposed path learned by Q-learning under epsilon-greedy exploration.
Open research repositoryAutomatic Image Captioning
CNN–LSTM pipeline using VGG16 and transfer learning on Flickr8k, improving BLEU score by 15%.
Open research repositoryRetinaFace Localization
Multi-task face detection with ResNet-152 and MobileNet, reducing localization loss by 40% across ten epochs.
Open research repository