Software Engineer | Software Security & Cyber-Physical Systems | Backend & Full-Stack Development
I am a software engineer working on enterprise and financial software systems, primarily using C#, ASP.NET Core, Angular, SQL Server, and REST APIs.
My research interests center on Software Security — especially the security of LLM agents and AI-enabled systems — with Cyber-Physical Systems (CPS) as a close second. My security focus is grounded in my day-to-day experience building secure, reliable enterprise software.
I also have academic research experience in machine learning and computer vision, including biometric systems and medical image analysis.
- Software Security
- Cyber-Physical Systems (CPS)
- Security of LLM Agents and AI Systems
- AI for Security
- AI for Software Engineering (AI4SE)
- Program Analysis and Automated Testing
- Secure and Reliable AI-Assisted Software Systems
- Software Engineering and Software Architecture
My professional work focuses on designing and developing enterprise software systems, including:
- Financial and transactional software
- Backend services and REST APIs
- Authentication and authorization systems
- Data reconciliation and processing systems
- Case management applications
- Database design and optimization
- Integration between enterprise applications
- Role-based access control and secure application workflows
C# Python SQL TypeScript JavaScript Java Go
ASP.NET Core ASP.NET MVC .NET Entity Framework Core
REST APIs FastAPI JWT Authentication
Angular HTML CSS JavaScript TypeScript
Microsoft SQL Server
Clean Architecture CQRS MediatR
Dependency Injection Git API Integration
Authentication & Authorization Role-Based Access Control
PyTorch Deep Learning Computer Vision
Transfer Learning Image Processing Local LLMs Ollama
LLM Tool Calling
A security gateway that sits between an LLM agent and the tools it can call, and decides — independently of the model — whether each tool call is something the user is allowed to do and actually asked for.
Built and evaluated on a local banking agent under prompt-injection attacks:
- A role-based check alone blocked only 3 of 8 attacks from a hijacked agent — it could not stop unauthorized transfers, because the user is allowed to transfer money.
- Adding task-scope and argument-provenance checks blocked 8 of 8, with no legitimate requests blocked.
Key areas:
- LLM agent security and indirect prompt injection
- Least privilege and task-scoped authorization
- ASP.NET Core gateway with audit logging
- Python agent on local LLMs (Ollama)
- Experimental security evaluation
A software security research project in two stages:
- Vulnerable target API — a deliberately insecure ASP.NET Core banking-style API with hand-planted vulnerabilities, built as a realistic test target.
- Roslyn security analyzer — a static analysis tool built on the .NET Compiler Platform (Roslyn) that aims to detect those vulnerability patterns directly in C# source code.
Key areas:
- ASP.NET Core Web API
- Secure coding and common web vulnerability classes
- Static program analysis
- Roslyn analyzers (C# syntax trees and semantic models)
The goal is to explore how automated analysis can catch security defects earlier in the development process.
An enterprise reconciliation platform designed for processing and comparing large financial datasets.
Key areas:
- ASP.NET Core Web API
- Angular
- SQL Server
- CQRS and MediatR
- Batch data processing
- JWT authentication
- Modular application architecture
Enterprise case management software designed to support structured workflows, reporting, and integration with external financial systems.
Key areas:
- Database schema design
- REST API development
- Role-based access control
- External API integration
- Reporting workflows
- Enterprise data management
An experimental full-stack AI assistant using local large language models.
Technologies:
- FastAPI
- Angular
- Ollama
- Local LLMs
- Python
The project explores integrating modern AI capabilities into conventional software applications.
A small 2D game project developed with Unity and published for the web.
Technologies:
- Unity
- C#
- WebGL
My academic research background includes machine learning, computer vision, and image processing.
Research involving iris image processing, segmentation, biometric recognition, and deep-learning-based approaches for non-ideal iris images.
Worked on deep-learning-based brain tumor classification using transfer learning and EfficientNet architectures.
My current interests focus on software security — including securing AI agents themselves — and on applying AI to software engineering problems such as vulnerability detection, code analysis, and testing.
I am particularly interested in questions such as:
Software Security
- How can LLM agents that call tools be kept within what the user actually asked for, even when the model is manipulated?
- How can static analysis catch security defects — including those in AI-generated code — before they reach production?
- How can AI assist vulnerability detection without overwhelming developers with false positives?
- How can authentication, authorization, and access-control flaws be identified in real-world web APIs?
Cyber-Physical Systems Security
- As AI agents are given control over physical systems, how can their actions be bounded by both authorization and safety constraints?
- How can malicious commands to industrial controllers be detected when they look like legitimate operations?
- How can false data injection and sensor spoofing be distinguished from ordinary faults and noise?
- How can software running on embedded and control systems be analyzed for vulnerabilities before deployment?
AI for Software Engineering
- How can AI improve software testing and debugging?
- How can we evaluate the correctness of AI-generated code?
I enjoy building reliable software systems and exploring how security, artificial intelligence, and software engineering can work together to create safer software.