Segu Venkata Chiranjeevi

VR Health Viewer

A full-stack web application that enables immersive 3D visualization of medical brain models (.nii files) using both traditional and VR (stereoscopic) viewing modes.

The application supports gesture-based interaction using MediaPipe and enables real-time processing through a Flask backend.

NLP-Based Goal Model Extraction

Classified software requirements into functional, nonfunctional, and ambient types using NLP and transformers (BERT, RoBERTa, DistBERT, ELECTRA).

Included preprocessing, class balancing, TF-IDF, and deep learning. Visualized results and performance evaluated.

Predicting Kubernetes Cluster Management Issues

Implemented and analyzed Kubernetes cluster deployment, service discovery, and networking issues.

Used K-means for anomaly detection and BERT for log interpretation to enhance fault resolution. Optimized LSTM-based sequential analysis for network traffic pattern recognition.

Weather Forecasting for New Delhi Using LSTM

Developed a time-series forecasting model to predict next-day temperatures in New Delhi using a stacked LSTM neural network trained on engineered weather features. Achieved a validation RMSE of 3.535°C and test RMSE of ~4.38°C.

Built a preprocessing pipeline for time features, lag variables, rolling statistics, and seasonal encoding. Trained and evaluated the model using 5-fold walk-forward validation, optimizing performance with dropout and Adam optimizer.

Forecasted next 5 days' temperatures and visualized results against actual data.

Data Analytics & Visualization using Power BI

Completed hands-on business intelligence tasks as part of a virtual internship focused on data-driven decision making using Power BI.

Developed interactive dashboards to visualize key metrics such as sales performance, customer segmentation, and product profitability. Cleaned, transformed, and modeled raw datasets using Power Query Editor and DAX for calculated columns and measures.

Created compelling visual reports using bar charts, slicers, KPIs, and trendlines to support business insights.

Image and Video Steganography System

A dual-mode steganography system enabling secure text and file embedding within images and videos using LSB and DWT/DCT techniques. Supports reliable data extraction with minimal visual distortion and optional AES encryption for enhanced security.

Image module uses LSB and DWT for secret message embedding in PNG, JPEG, and BMP formats. Video module extracts frames, embeds encrypted payloads, and reconstructs videos while preserving audio.

Performance evaluated using PSNR and SSIM metrics.

Real-Time Fraud Detection System

A hybrid system designed to detect fraudulent bank transactions using both static deep learning models and real-time streaming analysis via Apache Spark MLlib. Focused on benchmarking model accuracy, GPU utilization, and latency.

Implemented complete preprocessing pipeline including time-feature extraction, categorical encoding, and normalization. Compared LSTM-based models (Keras/PyTorch) with distributed PySpark ML models on cleaned transactional data.

Generated probabilistic fraud predictions and evaluated using PSNR, SSIM, and precision-recall metrics.

NLP Text Classification Pipeline

A comprehensive NLP pipeline that classifies text using both traditional ML and transformer-based models. Includes preprocessing, vectorization, model training, and evaluation.

Implemented TF-IDF, Word2Vec, and transformer embeddings (BERT, DistilBERT, DeBERTa). Trained Logistic Regression and Transformer models on labeled data.

Evaluated models using classification reports, confusion matrix, and ROC curves. Supports real-time predictions on custom text inputs.