At Innodata, I evaluated AI-generated responses for factual accuracy, reasoning, instruction adherence, safety, and overall quality across multimodal projects.
I also performed image annotation and editing quality assurance, and assessed vision-language models for object localization, mask accuracy, referring expressions, and pixel-level pattern recognition.
For my project on detoxifying code-mixed Tanglish, I developed an AI-based system and fine-tuned and evaluated GPT-2, LLaMA-2, and Mistral-3B to compare how well they generated safer text.
I also developed a MERN Stack hostel management application and an Android application connecting customers with tailoring services.

