The Business Challenge
The business situation the organisation faced before the project began.
Students often spend significant time searching through lengthy academic PDFs to find relevant information for examination preparation. Traditional keyword-based searching does not always provide meaningful answers. Students also struggle to organise information into answers suitable for different mark requirements. A system was needed to make academic content easier to retrieve and understand.
Why AI Was the Right Solution
Why AI was the right approach — and what alternatives were considered.
Academic documents contain large amounts of unstructured text, making manual searching time-consuming. Semantic search helps identify relevant passages based on the meaning of a question rather than exact keyword matches. A language model can then use the retrieved information to generate readable and structured answers.
How the Solution Was Delivered
Discovery, design, development, testing and rollout — the journey, not the tooling.
- 1Academic requirements analysis and planning
- 2Collection of four ISM unit PDFs
- 3PDF text extraction and preprocessing
- 4Text chunking and embedding generation
- 5Vector database construction and document retrieval
- 6Qwen language model integration
- 7FastAPI backend development
- 8Streamlit frontend development
- 9System testing and refinement
Key Technical & Architecture Decisions
Architecture, model selection, workflow and trade-offs.
Selected a Retrieval-Augmented Generation architecture to ground responses in academic documents. Used Sentence Transformers for semantic embeddings, ChromaDB for persistent vector storage in the local implementation, and Qwen2.5-1.5B-Instruct as the language model. FastAPI manages backend requests, while Streamlit provides the user interface.
Challenges & How They Were Solved
Obstacles hit along the way and how they were overcome.
One challenge was processing lengthy PDF documents into searchable content. This was addressed through text extraction and chunking. Another challenge was retrieving relevant passages for user questions, which was handled through semantic embeddings and vector similarity search. Model execution and GPU resource limitations also required additional testing.
User Adoption & Change Management
How users responded — training, change management and feedback loops.
The prototype was designed with a simple question-and-answer interface so students can interact with academic materials without needing technical knowledge. The system is intended for demonstration and evaluation, with further usability testing required before wider adoption.
Business Outcomes & Impact
The measurable outcomes recorded on this AI Case Study — with the story behind them.
The prototype demonstrates how AI-powered document retrieval can simplify access to academic information. It supports context-based question answering and structured exam preparation using course-specific materials. Quantitative improvements in study time, cost savings, and learning outcomes have not yet been measured.
Business Outcome Categories
The story behind the numbers
The prototype provides a centralised way to access information from academic materials. Students can ask questions in natural language instead of manually searching through multiple PDF documents. Retrieved passages help ground generated answers in the supplied course content. The project demonstrates the potential of RAG-based systems for educational knowledge management.
Lessons Learned
What surprised the team, what worked well, and what would be done differently.
- Document chunking directly affects retrieval quality.
- Semantic search improves context-based information retrieval.
- RAG helps ground language model responses in source documents.
- GPU availability affects local model inference performance.
- End-to-end testing is essential for reliable AI applications.
Future Opportunities
Where this solution could go next.
Future improvements include supporting additional subjects, improving retrieval accuracy, evaluating answer quality, adding source citations, optimising model inference, and deploying the application on a reliable hosting platform.
Professional Reflections
Guidance for another builder tackling a similar problem.
This project highlighted the importance of document preprocessing, retrieval quality, and reliable model integration in building RAG applications. It also demonstrated that a useful AI system requires careful testing of both retrieved information and generated responses.