Optimizing vision encoders for edge-deployed visual LLMs - ICCCNet 2026 (accepted; Springer LNNS). Benchmarked 8 architectures on 4 hardware platforms and found the vision encoder to be the main latency bottleneck. ICCCNet 2026 | published LNNS

• Benchmarked 8 architectures across 4 hardware platforms; TTFT ranging from 1.6ms to 98.0ms on identical hardware
• Identified vision encoders as the 70-85% latency bottleneck in edge VLLM deployments
• Introduced the Cache Cliff phenomenon and Sensory Bottleneck Hypothesis

• OpenTex - browser-based research-writing editor with LaTeX compile preview, local-first storage and venue templates (IEEE, ACM, Springer LNCS, arXiv).
• Diagflo - turns plain-language ideas into diagrams for product and engineering teams. Live demo
• GetCred - credit-intelligence prototype for gig workers with no credit history. Working prototype validated on simulated transaction streams; real pilot is the next step.
• Bitewise - food and health assistant with Gemini-based meal suggestions and local-only data.

• Founder's Office, Product and Engineering, PineBrook Technologies (2026)
• FinTech intern, Medikabazaar (2025)
• Tech Lead, Microsoft Learn Student Ambassadors chapter
• Maintained the MLSA org, contributing to repositories and chapter site
• KIIT MUN registration and event broadcast platforms kiitmun.org