Chisthia
Khan

A layered blue-team defence framework built to protect a local LLM (TinyLlama) from prompt injection, jailbreaks, system prompt leakage, obfuscated malicious input, and harmful cyber requests. Implemented as a modular Python/Flask pipeline.

• Implemented input sanitisation, obfuscation normalisation (Base64, leetspeak, spaced-character attacks)
• Developed rule-based risk scoring to classify prompts as benign, suspicious, or malicious
• Added safe-mode rewriting for borderline prompts and output validation to detect data leakage
• Used canary-token leakage detection to identify system prompt exposure
• Evaluated across an 80-prompt dataset; final layered defender achieved over 96% malicious block rate

Undergraduate thesis project applying DNA cryptography and AES-GCM encryption to secure sensitive medical image data. Connects theoretical cryptographic concepts with practical implementation in a privacy-critical domain.

• Applied DNA cryptography concepts to protect medical image data
• Implemented AES-GCM encryption logic for secure data protection
• Demonstrated sustained interest in applied cybersecurity research across academic career

Singapore-based invoice automation web application helping businesses streamline invoice handling, merchant operations, and administrative workflows. Delivered across five Agile development phases.

• Solely responsible for backend system development under team lead supervision
• Built features supporting invoice automation and merchant operation workflows
• Delivered production-level software over a long-term professional development cycle

Python client for the DS-Sim distributed systems simulator. Communicates with a simulated server environment and schedules jobs to appropriate servers based on capacity, availability, queue size, and resource requirements.

• Implemented socket-based client-server communication protocol
• Designed scheduling logic considering CPU, memory, disk capacity, server state, and queue size
• Tested across multiple XML-based simulation configurations using automated performance scripts

A farm inventory management web application for organising, managing, and tracking inventory data in a structured system built around real business-style operational requirements.

• Built full CRUD functionality for structured inventory data management
• Designed system architecture around practical operational workflows

An Android to-do and scheduling productivity application built with MVVM architecture, local SQLite storage, and scheduling/reminder functionality.

• Implemented MVVM architecture with ViewModel and LiveData
• Used SQLite for persistent local data storage
• Integrated AlarmManager/BroadcastReceiver for reminders and scheduling

A two-part Android healthcare application — a user-facing health services app and a separate admin panel for managing app data via Firebase. Demonstrates a complete application ecosystem with both end-user and admin functionality.

• Built user-facing healthcare app and separate admin management system
• Implemented Firebase-backed CRUD operations for admin control
• Designed clear separation between user-facing features and admin data management