Research prototype that analyzes smart contracts, blockchain activity, and anomaly signals to flag possible DeFi rug-pull risk, reaching 0.94 macro F1 on an 843-contract benchmark.

• Trained a fused multi-model classifier (tabular, Solidity source, opcode, and GRU timeline features) on 1,550 labelled contracts, reaching 0.94 macro F1 on an 843-contract benchmark across 6 rug-pull categories.
• Built a React/Vite and FastAPI workflow for contract address intake, feature extraction, and prediction review.
• Used Redis and Docker Compose to coordinate backend services and local experiment runs.

Limitations: Research prototype only; not production security tooling or financial advice. Public release needs stronger dataset provenance, license documentation, and exact model metrics.

IoT smart shoe prototype that collects force and motion data from an ESP32 device and visualizes live step, balance, and fall-risk signals.

• Collected pressure and motion readings from 3 force sensors (front, middle, heel) and an MPU6050, processed on ESP32 firmware with median filtering.
• Applied threshold and debounce logic to detect steps, falls, and uneven weight balance.
• Published readings through BLE notifications and MQTT into a live Next.js dashboard.

Limitations: Academic prototype only; not a medical or safety device. Results are sensitive to sensor placement, calibration, noise, and hardware reliability.

Computer-vision experiment using Food-101 and EfficientNetV2B2 transfer learning to test food recognition for a future nutrition-estimation pipeline.

• Trained an EfficientNetV2B2 transfer-learning model on the 101-class Food-101 dataset.
• Evaluated on 25,250 samples with 0.8493 accuracy and 0.8488 macro F1.
• Mapped the classification task to the first stage of a nutrition-estimation product path.

Limitations: Evaluated on 25,250 Food-101 samples with 0.8493 accuracy and 0.8488 macro F1; nutrition database integration, portion-size estimation, and app/API delivery are not complete.

Progressive web app for cultural knowledge exchange about Indian heritage.

• Built offline-capable web and mobile experiences.
• Collaborated across Mahidol University's ICT faculty and the Institute for Languages and Cultures of Asia.

Java OOP simulator for transactions and inventory management.

• Modelled transaction and inventory flows with object-oriented design.
• Practised applying design patterns in a small Java system.

Python implementation of RSA encryption and decryption for cryptography fundamentals.

• Implemented RSA encryption and decryption in Python.
• Practised core cryptography concepts through a focused assignment.