The Challenge
Acelot Innovation Private Limited needed a secure and scalable platform for conducting online examinations with real-time remote proctoring. The platform needed to provide students with a smooth examination experience while enabling administrators and proctors to monitor examination sessions and review potential proctoring events.
The solution needed to bring together several different workloads, including secure user authentication, examination management APIs, live video streaming, AI-powered video analysis, proctoring event generation, recorded examination videos, and application data management.
A major architectural challenge was that these workloads have very different resource requirements. Conventional API requests are relatively lightweight, while live video requires high bandwidth and AI-based video analysis can require significant compute and GPU resources.
Therefore, the platform needed an architecture where API, video-streaming, and AI-processing workloads could operate and scale independently, without allowing heavy video or AI workloads to negatively impact the core examination application.
The Solution
We designed a cloud-native AWS architecture that separates the platform into dedicated layers for authentication, application services, real-time video processing, AI proctoring, and data storage.
Application & API Layer
FastAPI was implemented as the primary backend and API layer, supporting both student and administrator operations. The API layer manages examination-related functionality, application requests, proctoring events, and communication with MongoDB.
The architecture keeps the API services as stateless as practical, allowing additional instances to be deployed as application demand increases.
Authentication & Access Control
FusionAuth was used as the identity provider for authentication and authorization. Administrators authenticate through FusionAuth and receive access tokens that are used to access protected application functionality.
Role-based access and token validation provide a foundation for protecting administrative and examination-related operations.
Real-Time Video Streaming
AWS IVS provides the live video streaming layer for remote examination monitoring. Student video is handled separately from normal application/API traffic, allowing the video workload to operate independently.
Recorded examination video is directed toward Amazon S3, providing scalable object storage for examination evidence and recordings.
AI-Powered Proctoring
The AI processing layer receives video frames and performs proctoring analysis. The resulting processed information and proctoring flags are sent back toward the application and data layers.
The overall processing flow is:
Student Camera → AWS IVS → Video Frames → AI Engine → Detection & Classification → Proctoring Flags → FastAPI / MongoDB
The architecture is designed so AI processing can evolve independently from the core application workload.
Data & Storage Layer
MongoDB is used for application and event data, including student, administrator, examination, and proctoring-related information.
Amazon S3 is used for static files and recorded examination videos. Keeping large video objects in object storage prevents the application database from becoming responsible for large media workloads.
Containerization & CI/CD
The platform is organized across dedicated repositories for the AI, candidate, backend/API, and administrator workloads.
Each repository has its own Jenkins CI/CD pipeline and Argo CD deployment configuration. Docker is used to package workloads, while Argo CD synchronizes the deployment configuration into the Amazon EKS environment.
The EKS cluster separates workloads according to their resource requirements, including a dedicated GPU-backed node for AI inference and separate infrastructure for the application workloads.
Security & Reliability
The architecture incorporates security considerations across identity, APIs, storage, databases, networking, and secrets management.
Recommended controls include secure authentication, token validation, HTTPS/TLS, private S3 buckets, least-privilege IAM, database access controls, restricted network access, centralized logging, backups, and controlled access to examination recordings.
The Results
The implemented architecture provides Acelot Innovation Private Limited with a modular foundation for an AI-powered online examination and remote-proctoring platform.
The separation of API services, live video streaming, and AI inference allows these workloads to be managed and scaled independently according to their individual resource requirements.
The solution also provides:
Scalable application and API architecture
Real-time examination video streaming
AI-powered proctoring analysis
Secure examination recording storage using Amazon S3
Structured proctoring event management using MongoDB
GPU-backed infrastructure for AI workloads
Independent CI/CD workflows for application components
Controlled Kubernetes-based deployment through Amazon EKS
Secure authentication and authorization
A foundation for monitoring, disaster recovery, and future AI enhancements
The architecture also creates a foundation for future capabilities such as real-time proctor dashboards, advanced examination analytics, automated retention policies, AI model evaluation, event-driven processing, and stronger disaster-recovery strategies.
Note: I have intentionally not added fake figures such as user counts, percentage improvements, uptime, latency, or cost savings. Your source document explicitly states that measured production performance figures were not provided.