AI-Powered PhD Exam Preparation Using Retrieval-Augmented Generation
KonkurRAG is an intelligent study assistant designed to help candidates prepare for the Iranian PhD entrance exam in Computer Science (AI). It leverages a RAG pipeline built on 10+ years of historical exam questions to provide precise, exam-aligned learning.
- Retrieves similar past exam questions
- Searches by topic, difficulty, or pattern
- Uses embeddings + reranking for high accuracy
- Generates clear walkthroughs for AI, ML, DSA, math, and reasoning questions
- Provides step-by-step logic aligned with exam style
- Detects weak topics automatically
- Recommends what to study next
- Builds targeted quizzes from real exam data
- Identifies repeated question types
- Highlights high-yield topics
- Tracks your progress over time
RAG Pipeline: PDF Exams → Text Chunking → Metadata Tagging → Embedding Model (E5/BGE) → Vector Store (Chroma/Weaviate/Qdrant) → Reranker → LLM (GPT / LLaMA) → Query Response Engine
konkurrag/ data/ pdfs/ (Raw exam PDFs) processed/ (Extracted & cleaned text) metadata.json (Tags: year, topic, difficulty)
src/ ingestion/ (PDF parsers, chunkers) embeddings/ (Embedding models & vector store) retriever/ (Similarity search + reranking) generator/ (LLM answer generator) pipeline.py (Full RAG pipeline) api/ (Optional REST API)
notebooks/ exploration.ipynb analysis.ipynb (Pattern mining & topic mapping)
README.md
Run the following:
git clone https://github.com/yourusername/konkurrag cd konkurrag pip install -r requirements.txt
Example: Run a query
from konkurrag.pipeline import ask answer = ask("Teach me AVL tree rotations with examples") print(answer)
Example: Retrieve similar past questions
ask("Show me past exam questions about Lagrange multipliers")
- Add exam difficulty prediction
- Add spaced-repetition flashcards
- Add UI dashboard (Streamlit)
- Add question auto-classifier
- Add progress analytics
Because the best teacher is 10 years of real exam data combined with LLM intelligence.
No randomness. No clutter. Pure exam-focused learning.
MIT License.