Sneha Singh

I built Tri Model, a multimodal system combining gaze, speech, and typing signals, with a FastAPI backend and synthetic data. Its Logistic Regression classifier reached F1 0.97 and AUC-ROC 0.9747 on that data; it has no clinical validation.

At Unified Mentor, I built end-to-end machine learning pipelines and benchmarked algorithms for classification and regression. I also implemented real-time inference endpoints and anomaly-detection modules, and deployed ML applications with Streamlit.

I built SADEV, a culturally adaptive mental health support system using a multi-stage LLM pipeline on the Claude API with an OpenAI API fallback. Its evaluation harness reports crisis recall above 0.90, macro F1 above 0.78, and Cohen's Kappa of at least 0.75.

At Adcamping India Private Limited, I clean, prepare, and analyze ongoing performance data using Python, SQL, and NoSQL databases, and build reports in Excel and Google Sheets. I also built Zia, an independent AI assistant modeled after Zomato's Zia, using a LangChain ReAct agent, hybrid RAG retrieval, and Redis session management.