ResNet Experts in Germany

• Developed end-to-end reproducible ML pipelines (PyTorch) with data versioning (DVC), experiment tracking (MLflow), automated testing (PyTest), and CI/CD across all training workflows.
• Scaled Vision Transformer and CNN training across NVIDIA A100 GPU clusters (CUDA, DDP, SLURM); applied hyperparameter optimization (W&B Sweeps) to reduce training overhead and identify optimal configurations.
• Developed a real-time 3D human motion generation system (ViT, VQ-VAE, SMPL-X/PIXIE) for personality-conditioned avatar synthesis; achieved state-of-the-art FID = 6.15 and P-FID = 10.31 on the UDIVA benchmark.
• Validated model expressiveness through structured user studies, achieving 86% accuracy in distinguishing extroverted vs. introverted avatar behaviors.
• Optimized inference pipelines by deploying PyTorch models via TensorRT and ONNX Runtime into native C++ code; benchmarked performance.

• Acted as technical consultant to optimize multi-layer ensemble models combining ResNet, CNN-BiGRU-Attention, and XGBoost.
• Guided implementation of a Logistic Regression meta-learner to solve class imbalance problems, achieving 92.86% accuracy and 0.9644 AUC on PTB-XL and Chapman-Shaoxing datasets.

• Designed and built a multi-tenant Python/Flask API platform with JWT + API-key authentication, scoped access control, and service-level orchestration as the backbone for AI applications.
• Built a multimodal RAG system with hybrid chunking, dense/sparse embeddings, hybrid retrieval, reranking, and vector search to deliver grounded, high-precision responses across enterprise data.
• Productionized AI workflows with Docker, CI/CD, Redis-backed async job tracking, webhook callbacks, external AI/media service integrations, and runtime health/reliability controls.

• Developing methods for monocular 3D facial reconstruction and personalized geometric modelling from mobile imagery
• Building learning-based approaches for facial shape estimation, video-based facial analysis, and privacy-preserving visual learning

• Developed, trained, and validated various computer vision models, including ResNet architectures for image classification as well as few-shot, two-shot, and first-shot detectors
• Used uncertainty modeling methods to improve robustness, reliability, and accuracy in real-world applications
• Increased model trustworthiness, especially in complex computer vision tasks

• Built a CRNN model with VGG16 and BiLSTM backbone for text-based CAPTCHA recognition
• Achieved 9.37% character error rate and 68.36% sequence accuracy on validation data
• Expanded data augmentation pipeline with distortions, noise injection, and clutter to improve robustness
• Conducted detailed error analysis on confusable characters (O, Q, D) and proposed error-specific augmentation
• Tech Stack: Python, TensorFlow/Keras, OpenCV, NumPy, Matplotlib

• Employed advanced natural language processing to analyze resume content, identifying key skills and experience relevant to specific job descriptions.
• Provided actionable recommendations for resume improvement, highlighting content gaps and suggesting optimization strategies.
• Evaluated and enhanced compatibility with Applicant Tracking Systems through keyword analysis and format optimization.

• Designed and implemented an advanced, agentic RAG pipeline using LangChain and LangGraph for structured data extraction from PDFs, utilizing tools, state management, and OpenAI LLMs (GPT-4) to improve accuracy and handle complex document structures.
• Developed a Google AI agent for extraction of structured information from PDF documents and deployed the agent on Vertex AI.
• Architected data pipelines using Azure Data Factory and Databricks to ingest data from Azure Blob Storage, process it with PySpark, and load it into Azure SQL Database via Linked Services.
• Containerized AI agents and services using Docker for consistent local development and deployment.
• Utilized PySpark and Dask for database querying in coordination with Azure Blob Storage and Document Storage.
• Created a ReAct agent that extracts structured data from PDF documents using tools and integrating Azure Document Intelligence.
• Contributed to the CPO invoices validation check project using Databricks to find existing CDRs and calculate total valid costs.
• Developed a conversational AI agent (chatbot) with a FastAPI backend, integrating RAG for precise tariff extraction and deployed the service using Azure Container Apps.
• Tools used: Azure, Azure OpenAI, Azure Document Intelligence, Azure Blob Storage, Google ADK, Google Cloud, Vertex AI, Gemini, Databricks, LangChain, LlamaIndex Ollama, Docker, PySpark, Azure SQL, Azure Data Factory, Azure AI Agent, Microsoft SQL Server

Detecting and classifying vehicles in images and video for traffic monitoring

• A YOLO + Faster R-CNN model built for real-world traffic and autonomous-vehicle scenarios. Awarded Best Paper Award at St Joseph Engineering College, March 2025.

What it does

• The model takes images or video frames and both localizes and classifies vehicles by type, making it usable for downstream applications such as traffic-flow monitoring or perception in autonomous-vehicle systems.

What I did

• Combined YOLO (for fast detection) with Faster R-CNN (for higher-precision classification), rather than relying on a single architecture, trading off speed and accuracy where each mattered most.
• Achieved 90% mean Average Precision (mAP), evaluated using IoU-based metrics rather than just raw accuracy, to properly reflect localization quality.
• Handled the full data processing and evaluation pipeline in Python using TensorFlow and OpenCV.
• The accompanying paper was awarded the Best Paper Award by the Department of CSE at St Joseph Engineering College.

Tech stack: Python, TensorFlow, OpenCV, YOLO, Faster R-CNN

• Tools: Maven, Jenkins, Docker, Sonarqube, Nexus and Spring Boot.
• Implementation and configuration of a CI/CD pipeline for a Spring Boot project.