Alternative Requirement Generator (RAG + FLAN-T5 / BERT)

This project is a prototype tool for generating creative software requirements using:

• Retrieval-Augmented Generation (RAG)
• Either FLAN-T5 (for full-text generation) or BERT (for masked word prediction)

Key Features

• Retrieval-Augmented Generation (RAG) using FAISS
• Sentence embedding with sentence-transformers
• Generative approach with FLAN-T5
• Masked Language Modeling with BERT
• Smart masking using spaCy (verbs and nouns)

How It Works

Step 1: Retrieval (RAG)

A small corpus of sample requirements is vectorized using all-MiniLM-L6-v2. Given a base requirement, we retrieve the top-k similar requirements using FAISS.

Step 2A: FLAN-T5 (Full Generation)

The base requirement + retrieved examples are passed to a FLAN-T5 model. The model generates creative alternatives via natural language generation.

Step 2B: BERT (Masked Prediction)

A key verb or noun in the base requirement is masked using spaCy. BERT then predicts possible words to fill in the masked position, generating creative variants.

Sample Corpus

The current prototype uses a small sample set of requirements. You can modify the array according to your need. In real-world use, this corpus would be much larger. But as this is a prototype so tried to keep it smaller.

Example Output

Base requirement:
"Users should be able to manage their subscriptions."

BERT Output: ( Only the masked word is predicted to generate new requirement)

1. Users should be able to renew their subscriptions.
2. Users should be able to update their subscriptions.
3. Users should be able to review their subscriptions.

FLAN-T5 Output: ( Full sentence is generated )

- Users should be able to customize subscription duration.
- Let users change their plan frequency anytime.
- Allow flexible billing options for subscribers.

Future works

• Use masked diffusion or LLM finetuning for higher creativity
• Expand to domain-specific requirement sets

This project is a simple prototype and it is inspired by research in Requirements Engineering, specifically:

• AI in Requirement generation