Ayushman Maurya

I’m an AI & Machine Learning student focused on building practical systems that perform in real conditions. I combine careful feature engineering with efficient data pipelines to get reliable results.

In multivariate time-series stock forecasting, I developed an LSTM-based model (PyTorch) that achieved 92% accuracy predicting 11 NIFTY-50 indicators. I built a sliding-window data pipeline for 5,000+ records, used Log-Normal transformation and StandardScaler to stabilize variance (reducing MSE by 18%), and leveraged CUDA plus learning-rate scheduling to speed convergence by 25% over a 150-epoch cycle.

In parallel, I engineered a RAG-enhanced AI agent using ChromaDB, NIM API, and SQLite. It delivers context-aware responses with average latency under 1.5 seconds, reaches 95% success in intent classification for administrative tasks, and improves semantic retrieval precision by 30% using vector embeddings over 1,000+ archived messages.

Outside the technical work, I’ve led community operations end-to-end—scaling a Discord community from 500 to 3,000+ members and automating moderation/event workflows with 10+ bots. I enjoy turning complexity into organized, measurable execution—whether that’s model training or community growth.