ConceptMap AI

Inspiration

Preparing for rigorous engineering entrance exams like JEE Main and AEEE is an exhausting marathon. I realized that taking mock tests and analyzing past year questions is only half the battle; the real friction lies in manually reverse-engineering mistakes to identify exact conceptual gaps. I wanted to build an intelligent engine that removes this friction entirely, transforming a single diagnostic test result into a high-yield, personalized micro-curriculum.

What it does

ConceptMap AI is a dynamic study engine designed to optimize learning efficiency. A student simply uploads a single diagnostic mock test result, and the platform instantly generates a targeted, 7-day study plan complete with a custom question bank. It maps out interconnected subjects, visually highlighting exactly which foundational topics need review before the student can tackle more complex problems.

How we built it

The core logic relies on an adaptive knowledge graph. We utilized Google AI Studio to process the diagnostic test inputs and generate semantic links between incorrect answers and their root concepts. The backend, hosted on Google Cloud, powers the generation of the custom study plans and targeted questions.

To model knowledge decay and logically prioritize which topics the student should study first, we integrated a spaced-repetition decay function into our backend routing:

$$ R(t) = c \cdot e^{- \frac{t}{S}} $$

Where \( R(t) \) is memory retention, \( S \) is the relative strength of the concept based on the student's test performance, and \( t \) is the time elapsed. This mathematical model ensures the engine always prioritizes the weakest links.

Challenges we ran into

The biggest challenge was tuning the AI to avoid hallucinating questions and ensuring the generated question bank perfectly matched the difficulty level of actual engineering exams. Establishing a seamless, low-latency connection between the uploaded document parser and the generative AI pipeline also required aggressive optimization of our cloud functions.

Accomplishments that we're proud of

I am incredibly proud of the single MVP flow. Taking a raw, messy diagnostic PDF and instantly outputting a clean, interactive, and highly actionable 7-day study plan feels like a genuine superpower for students.

What we learned

I deepened my understanding of prompt engineering for precise educational contexts and learned how to effectively deploy scalable AI microservices. This project also reinforced the importance of user experience; advanced backend logic means nothing if the resulting study plan is too overwhelming for a student to follow.

What's next for ConceptMap AI

The immediate next step is integrating a real-time feedback loop where the engine dynamically adjusts the 7-day plan as the student completes the custom question bank. We also plan to expand the platform's visual capabilities, allowing students to interactively traverse and expand their personal knowledge graph.

Built With

• fastapi
• generative-ai
• google-ai-studio
• google-cloud
• machine-learning
• python
• react
• tailwindcss
• typescript