Projects
Things I've built
Synapse
Lightweight Network Scanner
Network recon tools are either heavyweight black-boxes or lack composable, scriptable primitives for CIDR-scale scanning.
Built a stealth SYN scanner from raw sockets, parallelizing 1,024 port probes across CIDR subnets with Tokio. Wrote a binary DNS resolver from scratch per RFC 1035, parsing UDP responses with label-compression pointers for PTR lookups.
97 probes/sec on a single thread. Dual-stack IPv4/IPv6 ICMP host discovery via a blast-and-collect strategy with a 3-second sweep deadline.
PantryPal (AI-Powered Inventory & Recipe Generation)
Family members struggle to keep track of pantry items and plan meals efficiently.
Developed a React Native mobile app utilizing the Gemini 2.5 Pro API for nutritional label image recognition, integrated with Supabase Realtime for live data sync.
Achieved 95% accuracy on food label image recognition. Provided family members with easier inventory management and recipe suggestions, simplifying meal planning.
AutoHR (LLM-Driven HR Ticketing System)
Manual processing of HR requests sent via email is slow and requires excessive human effort for categorization.
Built a full-stack HR ticketing system using a Node.js backend to automate ticket creation from Gmail by integrating the IBM Granite 3.2 8B LLM via Ollama for AI summarization.
Automated ticket summarization and creation based on email content, significantly reducing initial processing time during the IBM Granite Hackathon.
Plant Disease Predictor (CNN)
Accurate and timely diagnosis of plant diseases is critical but often challenging and inaccessible to the average user.
Designed and trained a Convolutional Neural Network (CNN) with three convolutional layers for multi-label classification of plant leaf diseases using the Plant Pathology 2020 dataset.
Demonstrated proficiency in building production-ready deep learning models and utilizing CUDA for accelerated GPU training.
Crop Yield Predictor (Random Forest)
Predicting agricultural crop yields requires sophisticated modeling of multivariate data (weather, soil, historical data).
Created a machine learning pipeline using Random Forest models with extensive feature engineering, preprocessing, and GridSearchCV for hyperparameter tuning.
Achieved high predictive accuracy, validated using robust metrics like RMSE and R², demonstrating practical application of classical ML for economic forecasting.
Playground
Interactive experiments — click to launch