At IIT Mandi, I built AEGIS, a mobile application security analysis platform that screens Android and iOS apps against OWASP Mobile Top 10 rules and malware signatures. Its false-positive control pipeline reduced false positives from 35% to 5% while maintaining 90%+ recall.
On the Job Description Enhancer team project, I fine-tuned a language model with LoRA and DPO on job description pairs. I also worked on an automated judge loop that selects high-confidence pairs for training.
For the Political Advertisement Analytics Platform course project, I built a bilingual keyword classifier for political ad records and a dashboard served through a Next.js and PostgreSQL stack on AWS. I tuned database performance and built REST API endpoints for the analytics views.
As a volunteer on the VR Iris eye-tracking research project, I annotated iris and pupil regions across video frames and standardized the dataset for downstream machine learning. I also organized machine learning and deep learning workshops as a core member of KamandPrompt at IIT Mandi.
