Project Portfolio

Implemented advanced semantic segmentation with ResNet/VGG19 backbones achieving 98% pixel accuracy for vector-based design workflows. Used Label-Studio for annotation, Albumentations for augmentation, and developed RGB-to-SVG conversion pipeline processing 10k+ images with white rendering optimization.

Designed high-fidelity image generation system with VGG16 discriminator, reducing mode collapse by 30%. Implemented DCGAN architecture with LeakyReLU and Dropout layers, enabling users to dynamically synthesize high-quality images in under 2 seconds per iteration using Adam optimizer.

Created hybrid rule-based and ML system for e-learning platform boosting course completion by 20%. Engineered with scikit-learn and spaCy achieving 95% accuracy across 5k+ coding exercises. Integrated BERT-based sentiment analysis for feedback prioritization, improving learner satisfaction by 25%.

Built comprehensive financial analysis system synthesizing multi-decade datasets (2000-2025) with Revenue, EPS, and Operating Cash Flow metrics. Engineered hybrid data pipeline using Python and Pandas achieving 98% normalization accuracy. Implemented context-aware retrieval with LangChain and deployed interactive Streamlit dashboards reducing manual analysis time by 70%.

Deployed intelligent chatbot using historical data with NLTK for intent classification and entity extraction. Enhanced contextual understanding by fine-tuning BERT and integrating Word2Vec embeddings, improving conversational relevance by 35% (F1-score). Built end-to-end TensorFlow pipeline achieving <500ms latency per interaction.

Developed video processing model with hybrid VGG16/Conv2D architecture for facial expression analysis. Used TensorFlow and Albumentations for image augmentation with LabelMe annotation achieving 98% frame-level accuracy. REST API integration reduced latency by 25%, increasing customer satisfaction by 15-20% across partnered F&B outlets.

Built comprehensive sentiment analysis system using BERT fine-tuning and traditional ML approaches. Implemented data preprocessing with NLTK, feature extraction using TF-IDF and Word2Vec, and SQLite database integration. Achieved high accuracy sentiment scoring with multi-approach validation for customer feedback analysis.

Created self-driving car model with LSTM-CNN hybrid architecture for steering angle prediction from 70+ hours of dash-cam data. Synchronized multi-camera inputs via NVIDIA DRIVE PX and implemented YOLO for lane detection and object tracking. Optimized inference latency by 40% through TensorRT quantization, enabling 60 FPS decision-making on embedded systems.

Implemented sophisticated social connection prediction system using graph algorithms for 500k+ users. Utilized PageRank, Katz Centrality, Adar Index, and HITS algorithms with matrix factorization for latent relationship modeling. Trained Random Forest classifier achieving 92% F1-score with 88% precision and 85% recall in cross-validation.

Built recommendation system using visual similarity and NLP for fashion product discovery. Implemented VGG-19 neural network for visual product similarity and spaCy/NLTK for text processing. Used euclidean and weighted euclidean distance techniques with A/B testing to find optimal solution combining all results with business rules.

Comprehensive data analysis for restaurant expansion across Delhi, Bangalore, and Mumbai. Used Python, NumPy for statistical analysis and Seaborn/Matplotlib for visualization. Enabled quick informed decisions for restaurant opening and product selling, resulting in 28% sales improvement and 12% profit growth in initial month.

Investment opportunity analysis using comprehensive startup ecosystem data visualization. Identified clear funding trends and investment patterns using Plotly and Seaborn. Analysis showed seed funding and private equity as highest investment types, increasing chances by 40% for better funding deals in NCR & Bangalore and 30% improvement in investment raise success.