← Back to Home
Projects
Moments
A social platform for travelers to share 'sound-scenes' — photo + music moments.
TypeScriptNext.jsPostgreSQLVercel BlobRedisSpotify API
- Architected a full-stack social application with Next.js 15 App Router and Server Actions, implementing secure custom authentication and type-safe backend logic
- Integrated Spotify Web API and OpenStreetMap to create immersive 'sound-scenes' that combine visual memories with musical context and geolocation
- Implemented a scalable serverless infrastructure using Neon (PostgreSQL), Vercel Blob for media, and Upstash Redis for rate limiting and background job processing
Habit Tracker
A comprehensive digital notebook application for tracking daily habits, reflecting on progress, and planning long-term goals.
Next.jsReactSupabaseTailwind CSSTypeScript
- Built a responsive, production-ready journaling platform using Next.js and Tailwind CSS, featuring a custom book-style interface with theming support
- Integrated Supabase for secure user authentication and real-time database management of daily logs, habits, and goals
- Implemented interactive goal planning features with automated roadmap suggestions and visual habit tracking analytics
Arduino Solar Tracker
A dual-axis solar tracking system powered by an ESP32 that automatically aligns a solar panel with the sun for maximum light exposure.
ESP32C++ArduinoOLED DisplayServo Motors
- Designed and implemented a dual-axis tracking mechanism using four LDR sensors and two servo motors for real-time sun alignment
- Developed an adaptive control algorithm with light thresholds and noise buffering to ensure smooth and stable motion
- Integrated a 0.96” I2C OLED display to visualize live sensor readings and servo positions
- Optimized sensor data processing using buffered readings for improved accuracy under varying light conditions
NSFW Content-Filtering System
Modular content-filtering pipeline for text-to-video systems using DeepSeek and Phi-4
PythonDeepSeekPhi-4LLM EnsembleAI Safety
- Developed an NSFW content-filtering system for text-to-video prompts using Phi-4 and DeepSeek-R1.
- Evaluated 432 prompts, revealing a temporal risk detection gap — achieving 93% accuracy on non-temporal vs 61% on temporal risks.
Spectral Vision Transformer
Vision Transformer (ViT) model for hyperspectral image classification
PythonPyTorchSciPyScikit-Learn
- Designed a Spectral Vision Transformer for HSI classification on Indian Pines, achieving 95% accuracy through spectral-spatial patch embeddings and positional encoding