FlexAGE is a full-stack web application that tackles one of the biggest challenges in education: the time-consuming and inconsistent grading of subjective, open-ended questions. Developed as a prototype for USDC Global, this platform empowers educators to create rubric-based assessments and leverages Google’s Generative AI to provide fair, detailed, and near-instantaneous feedback, transforming a process that takes days into one that takes seconds.
The Problem: Scaling Subjective Feedback
While objective, multiple-choice questions are easily graded by computers, subjective work (like essays, reports, and open-ended answers) remains a manual, laborious process. For institutions dealing with tens of thousands of students, this presents critical challenges:
- Massive Time Sink: Educators spend days, not hours, manually grading, taking time away from teaching and curriculum development.
- Grading Inconsistency: Human fatigue and subjective bias can lead to unfair and inconsistent grading, especially across large student populations.
- Delayed Feedback: Students often wait weeks for feedback, long after the material is fresh in their minds, hindering the learning cycle.
The Solution: AI-Powered, Rubric-Grounded Evaluation
Under the expert mentorship of Dr. Raja Subramanian at USDC Global, I architected and built FlexAGE from the ground up. The solution provides a seamless, end-to-end platform for the entire assessment lifecycle.
- For Educators (Configurators): An intuitive UI allows them to create courses, design assessments with highly detailed, custom rubrics (defining specific criteria and performance levels), and enroll students.
- For the System: With a single click, the system sends a student’s submission and the corresponding rubric to an AI engine. The AI grades the work based strictly on the provided rubric, ensuring fairness and relevance.
- For Students: Students submit work and receive rich, engaging feedback in seconds, allowing for rapid iteration and a more dynamic learning experience.
Key Features & Their Impact
- AI Grading Engine: The core of the platform. By grounding the AI’s evaluation in a detailed rubric, FlexAGE reduces grading time from days to mere seconds, while internal testing showed the generated feedback to be encouraging and high-quality.
- Dynamic Rubric Creator: A powerful interface for building custom rubrics. This flexibility ensures the platform is not limited to one subject, but can be adapted for everything from literature to technical analysis. Data is stored efficiently in a PostgreSQL JSONB field.
- Secure Role-Based Access Control (RBAC): A robust three-tiered role system (
Student,Configurator,Admin) designed with FastAPI’s dependency injection ensures users only access what they’re supposed to. This was a key architectural challenge to ensure data privacy and system integrity. - Educator in the Loop: The system is designed to augment, not replace, the educator. An internal AI confidence score flags assessments that might need a human review, and configurators have the full ability to override and edit any AI-generated grade.
Technical Architecture Deep Dive
FlexAGE is built on a modern, robust, and scalable architecture designed for performance and maintainability, adhering to company standards at USDC Global.
- Frontend: A responsive UI built with Next.js 14 and TypeScript, aligning with team best practices. The interface is highly interactive, featuring role-specific layouts and a rich-text editor for submissions.
- Backend: A high-performance, asynchronous API built with FastAPI. Chosen for its raw performance, native async support for handling background AI jobs, and automatic API documentation which streamlined development.
- Database: PostgreSQL with SQLAlchemy ORM. The use of UUIDs for primary keys and JSONB for rubric data provides a secure and highly flexible data layer.
- AI Integration: A dedicated service class engineers and sends prompts to the Google Generative AI API. The prompt includes the student’s submission, the full rubric definition, and specific instructions, forcing the model to ground its entire reasoning process in the educator’s criteria.
- Deployment: Fully configured for serverless deployment on Vercel, enabling seamless CI/CD and infinite scalability.
Impact Points
- Architected and developed a full-stack, AI-powered assessment prototype to automate the grading of subjective work, reducing evaluation time from days to seconds for large student cohorts.
- Engineered a rubric-grounded prompting system for Google’s Generative AI, forcing the LLM to provide fair, consistent, and context-aware feedback, solving the key challenge of scalable subjective assessment.
- Designed and implemented a secure, asynchronous backend using FastAPI and PostgreSQL, featuring a three-tiered Role-Based Access Control (RBAC) system and a JSONB-based schema for dynamic, user-created rubrics.
- Built a responsive, role-specific frontend with Next.js 14, TypeScript, and Tailwind CSS, providing a seamless user experience for students, administrators, and course configurators.
- Managed the full project lifecycle from conceptualization to a functional prototype, demonstrating end-to-end ownership and the ability to translate business needs into a technical solution.