Secure AI Workflow in Bi-directional Client-Server Architecture

Project overview

This internship project focuses on designing and implementing secure AI-based processing of multimodal submissions, with controlled handling of user data before and during execution of an AI pipeline. The objective is to improve the privacy and security of AI model processing on the REDX platform while ensuring that sensitive content is handled safely throughout the pipeline and AI decision workflow with minimum loss of information.

The platform already processes user interactions, operational outcomes and AI-assisted processing results. As these flows evolve, stronger privacy protection and secure operation are required.

Project goals

• Design privacy-preserving, secure and encrypted AI inference workflows for the REDX platform.
• Investigate Trusted Execution Environments (TEE) and confidential computing.
• Assess the feasibility and performance trade-offs of Secure Multi-Party Computation (MPC) and split inference for multimodal workloads.
• Investigate the performance, security level and latency of the developed solution.
• Apply execution-acceleration techniques to reduce end-to-end inference time.
• Protect AI decisions against prompt injection and manipulation attempts to ensure reliable and trustworthy inference.

Required skills

• Knowledge of data engineering and data quality, with experience in Python and backend integration.
• Understanding of AI/ML, multimodal models and inference pipelines.
• Awareness of security, privacy-preserving techniques and access control.
• Analytical skills and English communication skills.
• Familiarity with Git.

Deliverables

• Technical report detailing the platform design, implementation and evaluation.
• Source code and documentation for the secure AI workflow-processing solution and prompt-injection prevention mechanism.
• Performance metrics covering data quality, AI inference performance, security and safeguarding efficiency.
• Demonstration prototype showcasing secure AI processing.

Conditions

• Recommended period: 4 to 6 months.
• Monthly stipend, with a possible end-of-internship performance bonus.
• Possibility to work with end customers.