Submission Instructions
Submit the following:
Required Files
1. Source Code
2. Final Report (PDF)
3. Blog Post
4. Github Link
5. Screenshots/Results
File Format
Compress all files into one ZIP archive.
Save using:
[Link]
Example:
[Link]
(Use same format exactly.)
Not Allowed
The following are prohibited:
Copying complete GitHub repositories
Using ready-made RAG templates without explanation
Pure prompt-only chatbot without retrieval
Plagiarism or copied reports
Violation may lead to zero marks.
Implementation and Evaluation of a Basic
Retrieval-Augmented Generation (RAG) System
Project Objective
The objective of this project is to design and implement a basic Retrieval-
Augmented Generation (RAG) system that can answer user questions using
an external document collection.
Students will apply concepts of:
Document preprocessing and chunking
Information retrieval
Dense embeddings
Retrieval-augmented response generation
Experimental evaluation
The project should demonstrate understanding of both the retrieval and
generation components of RAG.
Problem Statement
Build a question-answering system that:
1. Accepts a user query.
2. Retrieves relevant document chunks from a custom knowledge base.
3. Uses retrieved context to generate an answer.
4. Evaluates system performance on sample questions.
Suggested Domains (Choose One)
Students may select one domain:
Medical FAQ Assistant
Legal Document Question Answering
Research Paper Assistant
(Use at least 8 documents or an equivalent textual corpus.)
Part 1 - Corpus Preparation and Chunking (10 Marks)
Deliverables
Raw documents folder
Preprocessed corpus
Chunked data
Part 2 - Embeddings and Retrieval Module (20 Marks)
Implement retrieval using the following:
TF-IDF Retrieval
Dense Embedding Retrieval
Sentence Transformers
The system should retrieve the top-k relevant chunks.
Retrieval Comparison (Required)
Compare:
Classical retrieval (TF-IDF or keyword)
Embedding-based retrieval
Use at least 8 test queries.
Part 3 - Generation Module (20 Marks)
Use retrieved context to generate answers.
Open Source LLM
For each query show:
User question
Retrieved context
Final generated answer
Part 4 - User Interface (10 Marks)
Develop a simple interface using one:
Streamlit
Gradio
The interface must allow:
User input question
Answer generation
Display of retrieved evidence
Part 5 - Experimental Evaluation (10 Marks)
Evaluate system using at least 8 questions.
Include:
Correct responses
Failure cases
Observed limitations
Part 6 - Project Report (15 Marks)
Report should include:
1. Introduction
2. Methodology
3. System Architecture
4. Experimental Results
5. Discussion
6. Conclusion
Recommended length: 8–12 pages
Part 7 - Viva (15 Marks)
Bonus (Optional +5 Marks)
Any one enhancement:
Source citation highlighting
Chat history memory
Hallucination reduction strategy
Improved reranking strategy