At FlyRankAI, I built an unsupervised clustering pipeline that grouped published content pages into three performance archetypes using Google Search Console data. I validated the clusters with silhouette score, inertia, and comparisons against GMM and hierarchical clustering.
I tested the pipeline on anonymized data from January through June 2026 using a time-based, page-level train/test split. Its results agreed with an independently built rule-based review system 64.66% of the time.
For the AI Real time Interview Multi-Modal Agent project, I designed the backend architecture and reduced token cost by 80% through compression and caching. I also built the meeting interface and deployed the backend infrastructure on Google Cloud Platform.
I designed and deployed Chatbot-AKSBOT-AI, a voice-and-text chatbot selected as Best Overall Project at the GDG Hackathon. Other projects include a fetal brain abnormality risk assessment application and Neonet, a neural network library implemented from scratch with NumPy.
