Pranav Thorat

Built and deployed a neuro-symbolic AI decision-automation platform that automates compliance, cybersecurity, and operational workflows, cutting deviation response times by 80%.

Developed a dropout risk model processing over 10M records, improving retention by 15% and creating a user interface that reduced help-desk tickets.

Developed a student dropout risk model processing 10M+ enrollment records using Snowpark and Snowflake. Engineered 100+ features with PCA and autoencoder dimensionality reduction. Built an ensemble model combining XGBoost, logistic regression, and deep learning with 20% improved accuracy. Optimized hyperparameters via Bayesian optimization and ensured interpretability with SHAP values, increasing student retention rates by 15%. Created a domain-specific RAG engine for NYU courseware utilizing LangChain, FAISS, and LoRA-tuned Transformers. Deployed AI Q&A microservice on Kubernetes with Airflow and MLflow for versioned training and drift retriggers, achieving 25% accuracy lift at 300 ms p99 latency.