All Projects
Everything I've Built
9 projects across full-stack, mobile, ML systems, and security.
Showing 9 of 9 projects
Streetwise: NYC Rental Risk Scoring Tool
Node.js • Express • MongoDB • NYC Open Data • Vitest
A rental risk scoring tool for any NYC address, built for the NYC (NYPL x MLH) Hackathon 2026. Scores Building Health and Block Quality against precomputed citywide baselines using NYC 311 complaint data, with a provider-agnostic AI layer generating plain-language explanations.
Tap to see full details →
URSA: Real-time Video Surveillance & Analytics
NVIDIA DeepStream • Apache Kafka • Apache Spark • Flask
A real-time video surveillance and analytics platform sustaining ~30 FPS per stream across up to 6 concurrent 1080p feeds, led by Vishal as a 3-person team. 2nd place at Anveshan National Student Research Convention (2023) and published in Springer (LNDECT), 2024.
Tap to see full details →
VenomFlow: Preference Data Poisoning Research
TinyLlama • DPO • trl • Python
A research pipeline investigating preference-data poisoning in LLM alignment, built with a 5-person team at Virginia Tech. Served as integrator unifying the team's modules, implemented a DPO fine-tuning workflow for TinyLlama, and built the Attack Success Rate evaluation pipeline behind the paper's published results.
Tap to see full details →
AI-Driven Adaptive Traffic Management
YOLOv5 • FastAPI • Computer Vision
A real-time traffic signal optimization system that uses YOLOv5 to detect and count vehicles in live video streams, then dynamically adjusts signal timings based on vehicle density, reaching 94% detection accuracy on benchmark datasets.
Tap to see full details →
LunaFlow: Cycle-Aware Productivity Planner
Express • TypeScript • Prisma • React • Vite • shadcn/ui
A cycle-aware productivity planner that syncs with Google Calendar and Google Tasks. Built the Google Calendar and Tasks integration end to end and the color-coded cycle-phase calendar component, on an Express/TypeScript/Prisma backend and a Vite/React/TypeScript/shadcn-ui frontend.
Tap to see full details →
Sentimint: Multi-Agent Stock Analysis Tool
LangChain • Next.js • Python
Built the financial data scraping package behind a multi-agent stock analysis tool, exposing a tested, modular scraper through a LangChain Tool interface. Also built the Next.js frontend querying the pipeline and surfacing its output.
Tap to see full details →
CareRoute: Medical Tourism Travel Planning App
Flutter • Riverpod • go_router
A Flutter patient-booking app for a medical tourism travel-planning system, built for Marriott Codefest 2025 at Virginia Tech. Integrates with a teammate-built multi-agent backend orchestrator over REST for booking creation and voice-call triggering.
Tap to see full details →
No Roll Contendere
Static Analysis • AST • Python
An AST-based static analyzer for PyCryptodome API misuse, tracing hardcoded keys and salts across function boundaries through interprocedural call-graph analysis, to detect insecure AES modes, missing IVs, and unsalted PBKDF2 calls.
Tap to see full details →
Toxic Comment Detection
BERT • TensorFlow • Multi-label Classification
A multi-label toxic comment classifier fine-tuned on BERT with TensorFlow, reaching 92.3% test accuracy across 6 toxicity categories: toxic, severe toxic, obscene, threat, insult, and identity hate.
Tap to see full details →