AI-Driven Biology Practice Exam System

I need someone to build a prototype that I can use for experimenting with a biology class. Here's the requirements: (link removed) it'll require front end, back end experience + ability to work with OpenAI APIs. Details below: AI-powered practice exam system This is a prototype for a practice exam system that will be used in a university-level introductory biology course with around 800 students. The goal is to evaluate whether an AI-powered guidance system can improve student learning outcomes on the actual midterm exams. This is a functional prototype, so the user interface does not need to be polished. The focus is on the backend functionality to support the university experiment. All data: questions, learning objectives, textbook references will be provided in CSV files. I'm looking for a fixed quote as this is a student/project experiment. We're not building anything to move to production to test on other users other than our 800 students. I also need this to be production ready for the 800 students by end of the year (12/31) as I need it for January's semester. It doesn't have to be pretty but it needs to be functional. I expect OpenAI GPT4o or GPT4o mini to be used. API keys will be provided. Database Model: Student Table Name (from Google account) Email (from Google account) Questions Table Question ID Question Text Choice A, B, C, D, E (multiple choice options) Correct Answer Learning Objective ID List (one Question ID can reference multiple Learning Objective IDs) Textbook Sections Has Image (True/False) Image URL Learning Objectives Table Learning Objective ID Title Summary Textbook Page References Exams Table Exam ID List of Learning Objective IDs (10-15 per exam) Exam Questions Table (cross reference between Exam and Questions table) Exam ID Question ID Condition (1, 2 or 3) Exam Students Table (cross reference between Exams and Students) Exam ID Student ID Transcript (BLOB) Exam Question Students Table (child of Exam Students Table) Exam Student ID FK Question ID Answer Boolean Correct/Incorrect System Functionality: Students log in using their UC Davis Google accounts. For each exam, 10-15 learning objectives are provided. For each learning objective, 1-2 questions are randomly selected from the Questions table (the questions + learning objectives are all provided in a CSV file). The questions are assigned to one of two conditions: Condition 1: Standard multiple choice with right/wrong feedback Condition 2: Guided resource (provide relevant textbook sections) The conditions are randomly assigned to the questions. Students complete the practice exam, with their responses and the provided guidance/feedback stored in the Exam Students table. After the practice exam, students take the actual midterm exam. Their performance on the learning objectives covered in the practice exam is compared to their performance on the uncovered learning objectives. The complete transcript of each student's practice exam interactions is stored, including the questions they saw, their responses, and the feedback provided. This allows the instructor to review individual student data if needed. The system uses the existing bank of questions and learning objectives. No new question generation is required. If a question has an associated image, the image URL is displayed to the student during the practice exam. Image URL will be provided so the GPT API can pull up the referencing image. The system is hosted on Google Cloud Platform (GCP) using Firebase for authentication and database storage. We can also use AWS but we need Google account authentication. How is AI used? The idea is to leverage the conversational and question-answering capabilities of advanced LLMs to provide students with tailored support during the practice exams, beyond just static text-based hints. The LLM just needs to be 100% aware of the context in the CSV files which shows question association with learning objectives which gets associated with textbook references/summaries Specifically: For the "Guided Resource" questions (Condition 2), instead of just providing a link to the relevant textbook section, the system will use an LLM to engage the student in a conversational dialogue. If the student answers the question incorrectly, the LLM will analyze their response, identify the areas of confusion, and provide personalized guidance. This could include: Clarifying concepts Directing the student to re-read specific parts of the textbook Asking probing questions to help the student arrive at the correct understanding The LLM's responses will be dynamically generated based on the student's input, rather than a predefined set of hints. For the standard multiple-choice questions (Condition 1), the LLM can be used to generate additional explanation or context around the correct answer, helping the student build a more comprehensive understanding. By incorporating these LLM-powered features, the goal is to create a more engaging, adaptive, and learning-focused practice exam experience for the students, with the potential to positively impact their performance on the actual midterm exams. Some implementation necessities - A data structure with relational data (users, questions, L.O.s, exam, exam questions, exam students) - Login and user creation functionality with Google OAuth - A user-facing website with exam taking functionality and an optional admin-facing app with review functionality - AI features for feedback w/ prompt engineering - Possible vector storage database setup for textbook and exam context for reference/summaries. We need to store that student X, did exam Y with questions 1,2,3. Also that question 1 was in condition 1, question 2 was in condition 2, question 3 was in condition 2.... - Integration and transformation of existing question bank - Devops with deploying code to cloud We'll need a few screens: 1. Login 2. Exam selection 3. Exam-taking (variations: MCQ w/ Image, MCQ w/o image, guided resource, answer feedback) 4. Pre- and Post-exam screen w/ instructions 5. Instructor admin panel for me to set that Exam 1 should use learning objective XYZ (if a UI takes too long, I just need someone to show me how to update the db with this info) We don't need a practice vs midterm, they will all be practice exams. Real midterms are not through this tool and it's in person on paper, that's not in scope. This whole app is a website, so not an android/google app or anything, just a website.