At Handshake AI's Octahedron Project, I compare AI personal assistants on real-world tasks and score their responses for correctness, coherence, helpfulness, and tone. I write evidence-based justifications and create everyday Bengali prompts to assess how naturally and appropriately the models respond.
At Outlier, I evaluate Bengali voice-model conversations and text-to-speech audio, using project rubrics to assess response quality, pronunciation, and naturalness. I also review egocentric video captions for robotics training data, checking clip segmentation and correcting annotation errors against project guidelines.
Previously, I worked at Tabsons India as a Media Logger and then a Media Quality Analyst for Bengali and Assamese news channels. I logged and tagged broadcasts, then reviewed media and metadata for accuracy, completeness, and compliance using Xentaurix and RBM.

