← All Projects

Microsoft — Social Sentiment Pipeline

Multi-source sentiment pipeline on 1,000 images + 240 videos; 4 LLMs benchmarking vs RoBERTa; +40% accuracy, 1–2 h saved per cycle.

PySpark/Databricks pipeline, auto-translate to widen dataset, contextual analysis via 4 distinct LLMs & small LMs handling emoji/mixed-mode sentiment. Code sample available.

View related experience →
Agentic AIAI DeliveryData Platform