Aryakumar Mishra

I am an AI and machine-learning-focused engineer specializing in multimodal systems, transformer-based NLP, and production-ready model deployment. I combine strong academic performance in Electronics and Computer Science with hands-on research and product experience.

My recent AI research internship produced a sentiment-augmented fake-news detection model using BERT and RoBERTa, improving validation F1 to 0.70 and boosting recall on noisy datasets via a feature-fusion pipeline. I emphasize evaluation for class imbalance and real-world robustness.

I have built end-to-end projects including a multimodal DeepFake detector (>90% accuracy) using EfficientNet, Xception, ConvNeXT, Swin, ViT and optimized training throughput, plus a RAG chatbot with OCR, ASR, FAISS and Llama-3.8B for unified PDF/image/audio querying.

I deliver production-capable systems using FastAPI, Streamlit, Flask and React, and apply MLOps practices with Docker, CI/CD, MLflow and DVC. I enjoy solving practical problems, scaling models, and collaborating cross-functionally to take research into deployable products.