Consider this:
It’s 3 days before your exam.
You have 5 units. 2 are huge. 1 is confusing.
And you’re thinking:“What do I actually study?”
So you open past papers.
You start spotting patterns… maybe.
But it’s slow. Inconsistent. And honestly — a bit of guessing.
💡 What if that entire process was automated?
What if you could:
- Upload past papers 📄
- Instantly see what matters most
- Identify high-weightage topics
- Know what’s missing from your prep
- Get a day-wise study plan
And that's the solution I built and prototyped in 6 hours during a GenAI hackathon
🎯 Introducing: AI Exam Strategist
A system that analyzes past question papers and turns them into actionable study strategy.
Not just summaries. Not just answers.
👉 Actual decision-making support for exams.
🧠 What It Does
📂 Multi-Paper Analysis
Upload multiple past papers → the system processes them together and extracts meaningful patterns.
🔍 Pattern Detection
- Finds frequently asked topics
- Classifies difficulty levels
- Identifies year-wise trends
👉 Helps you focus on topics with the highest exam impact
📚 Syllabus Mapping
Upload your syllabus → instantly see:
- ✅ Topics already appearing in exams
- ❌ Topics not covered (potential blind spots)
📊 Visual Insights
- Topic frequency charts
- Difficulty distribution
- Topic vs difficulty breakdown
- Year-wise trends
👉 Patterns become obvious at a glance
🧠 Smart Study Planner
Generates a day-wise plan based on:
- available time
- topic importance
👉 Designed for maximum ROI under time constraints
📝 Practice Question Generator
Select a topic → generate relevant practice questions instantly.
💬 AI Assistant
Ask:
“What should I prioritize?”
Get answers grounded in your own analyzed data.
🏗️ Tech Stack
- FastAPI → backend APIs
- Streamlit → interactive UI
- Groq API (LLM) → classification & generation
- LangGraph → structured workflow orchestration
- Pandas → data processing
⚙️ How It Works
Upload Papers → Extract Questions → Classify (Topic + Difficulty)
→ Analyze Patterns → Map with Syllabus → Generate Insights
→ Create Study Plan → Practice + AI Chat
🤔 Did I Use RAG?
Not in this version.
Since the dataset is relatively small, I used:
👉 context injection (passing structured analysis directly to the LLM)
This keeps the system fast and simple.
For larger-scale usage, this can evolve into a RAG-based system with vector search.
📏 Evaluation (Keeping It Real)
I added a basic evaluation layer to understand how the system behaves.
- Used a small, manually created dataset
-
Measured:
- Topic classification
- Difficulty classification
⚠️ Important:
- Accuracy may appear low if you try it yourself
-
Because:
- dataset is small
- matching is strict (semantic matches may be marked wrong)
👉 The goal wasn’t perfect scoring —
but to validate the system’s reasoning and consistency
🧠 What I Learned
- Building GenAI systems is more about pipelines than prompts
- LLM outputs are messy — normalization is critical
- Evaluation in AI is not straightforward
- Simple approaches (like context injection) can outperform complex ones for MVPs
- Speed + clarity > overengineering
🔮 Future Improvements
- OCR for scanned PDFs
- Semantic topic matching using embeddings
- Persistent memory across sessions
- Scalable deployment
🎥 Demo & Links
- 🔗 GitHub: https://github.com/Anucool419/AI-Exam_Strategist
- 🎥 Demo Video: https://www.loom.com/share/04005565701e45d1855d1fa13bcee73a
- 🌐 Live App: https://ai-examstrategist-ryjarq6usrfbsd85gipexy.streamlit.app/
🏁 Final Thoughts
Exams aren’t just about how much you study.
They’re about:
- what you choose to study
- how you prioritize
- how well you use limited time
And right now, students are expected to figure that out manually.
This project explores a simple idea:
What if AI could guide those decisions?
Not replace studying.
Not shortcut learning.
But make preparation more focused, more intentional, and more efficient.
Because sometimes, the smartest move…
is knowing what not to study.
In ~6 hours, this went from an idea to a working system.
It’s not perfect — but it solves a real problem:
Maximizing study ROI when time is limited
Would love your thoughts 👇
Top comments (4)
Nice work👍
Thank you Ashrut!
Keep up the good work!
Sure 😀