Hi, I’m Arnab — a systems security engineer working at the intersection of AI evaluation, sandboxed execution, and secure infrastructure. My work focuses on building production grade safety controls for environments where untrusted or model generated code must be executed under strict security boundaries.
I have 10 years of experience across enterprise systems, utility platforms, cloud environments, and AI safety tooling, with a strong emphasis on policy enforced execution, runtime observability, and secure inter process communication.
I am an active open source contributor, with merged contributions to Inspect AI (UK Government / AISI) and ongoing work across LLM runtime safety ecosystems.
- Secure execution environments for AI evaluation and red teaming
- Policy enforced sandboxing and isolation for model generated code
- Runtime safety and observability for LLM serving systems
- Post Quantum Cryptography applied to real world IPC and infrastructure
- Maintainer friendly, opt in security extensions for open source projects
- Sandbox architectures and policy enforced execution
- Secure IPC using Unix sockets with modern and post quantum cryptography
- Runtime observability, audit logging, and security event tracing
- HTTPS and transport security for legacy systems
- Custom Linux kernel configuration and compilation for ARM based devices
- Kernel customization for device branding and enforcement of system level security policies
- Secure execution environments on embedded and edge platforms
- Linux systems programming across x86 and ARM
- Production deployment of NIST standardized PQC algorithms
- Dilithium for digital signatures
- Kyber for key encapsulation mechanisms
- Practical integration of PQC into secure IPC and industrial systems
- Performance conscious cryptography for constrained environments
- AI evaluation pipelines and sandboxed execution
- vLLM deployment and runtime safety improvements
- Multimodal and vision model deployment on ARM based devices
- Model optimization including int8 quantization and latency reduction
- Standalone policy enforced sandbox extension for safe model execution
- PyPI package published and documentation merged upstream
- Opt in GPU memory warning system to prevent out of memory crashes
- Feature is fully additive and opt in (PR in progress)
- Privacy first, opt in feedback extension for scanner evaluation
- No changes to existing scanning logic or public APIs (PR in progress)
I’m interested in collaborating on:
- AI evaluation safety and sandboxing
- LLM runtime security and observability
- Policy driven execution controls
- Post quantum cryptography in real systems
- Security focused systems programming
- Maintainer friendly open source extensions
- 📧 Email: papan.mitra.2121@gmail.com
- 🔗 LinkedIn: https://linkedin.com/in/arnab-mitra-6014a6a3