Hello, I’mSushant Kothari.

Autonomous multi-step document processing agent engineered with Python, LangChain, and OpenAI/LLM tool calling for dynamic query decomposition, structured JSON extraction, and autonomous agentic workflows across complex PDF & text documents.

Autonomous multi-agent research pipeline using LLMs, real-time web searching, and automated synthesis to generate deep corporate intelligence and technical landscape reports.

Production-grade Bio_ClinicalBERT pipeline leveraging long-document chunking and logit aggregation for scalable, accurate medical-specialty routing with reduced manual triage and GPU cost.

GPU-accelerated deep-learning pipeline for brain MRI tumor classification and pixel-level segmentation, enabling fast, reproducible evaluation and reliable medical-imaging research.

End-to-end human activity recognition system using a hybrid TCN-Transformer to classify 12 wearable-sensor activities for reliable digital-health monitoring, rehabilitation insights, and real-time inference.

Production-grade hybrid Transformer-inspired ensemble combining self-attention, CNN feature extraction, and gated-MLP fusion for cardiovascular risk prediction with robust evaluation and deployment-ready inference.

Real-time multilingual translation system using fine-tuned Transformer models, synthetic data augmentation, and microservices for enterprise-grade localization without external translation APIs.

Multimodal injury risk classifier combining dilated TCN, BiGRU, and Transformer self-attention to model wearable time-series data for accurate injury prediction beyond traditional workload metrics.

Vision system combining ConvNeXt and dual-attention refinement for robust infrared gesture classification with cross-subject generalization for AR/VR, assistive technology, and touchless interfaces.

Production-grade Transformer-based misinformation classifier using distilled pre-trained models, stratified training, and robustness validation for efficient, scalable content moderation.