Piyush Singh

I build end-to-end AI applications that turn machine learning concepts into practical products. My work includes a research paper analyzer that lets users query PDFs with source-grounded answers and page-level citations.

For the analyzer, I built a RAG pipeline using arXiv ingestion, PyMuPDF extraction, custom PDF chunking, SentenceTransformers embeddings, cosine-similarity search, LLaMA 3.1 through the Groq REST API, and a deployed Streamlit interface. I also developed a CNN image classifier that reached 96% test accuracy through augmentation, dropout, careful dataset splitting, hyperparameter tuning, and early stopping.

I've also built a content-based movie recommender and an IoT smart vehicle anomaly detection device that sends real-time alerts. I presented the vehicle project to the IIT Patna Incubation Center, where it was recognized for real-world problem solving.

Outside of projects, I run a YouTube channel with 30,000+ subscribers and 5M+ views, generating INR 3,00,000+ through data-driven content strategy. I bring the same focus on experimentation, audience analytics, and execution to the AI products I build.