entropy

Developed a regression framework to predict per capita income across 346 districts using heterogeneous datasets. Standardized messy categories via fuzzymatching pipelines, eliminating schema drift and ensuring dataset consistency. Benchmarked multiple models, identifying XGBoost as superior to a Linear Regression baseline, with R² of 0.75 on held-out test set. Leveraged constrained optimization on XGBoost predictions to impute missing per capita income values across 102 districts.

Engineered a LiDAR AR pipeline converting scenes into metrically-accurate OBJ meshes with ARKit, RealityKit, Model I/O. Implemented scene understanding with occlusion, physics; aggressive render optimizations deliver stable, low-latency visualization. Transformed vertices to world-space; fused multi-anchors into unified MDLAsset, exporting timestamped OBJ artifacts. Deployed robust networking: multipart uploads, geotagging, Codable responses, hardened error semantics with feedback.

Built an AI-driven geospatial tool to automate zoning data extraction, reducing manual analysis time by 40%. Used spatial clustering and pattern recognition to identify potential zoning issues & inform planning decisions in overlay areas. Leveraged NLP models to analyze application FAQs, automating responses & reducing processing time for BHI teams and clients.