SKSharmila K.A.

Sharmila K.A.

*This is an algorithmically generated author record based on publicly available information.

Sharmila K.A. has a total of 1 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Towards a SMARTer Requirements: An AI-based Approach for Requirements Quality

    by Sharmila K.A., Ramakrishnan Raman

    It is well recognized that the quality of textual requirements plays a critical role in engineering of complex systems. When textual requirements suffer from issues such as ambiguity, inconsistency, and lack of verifiability, it causes significant issues in the systems engineering lifecycle of complex systems. This paper presents the design, development and fine-tuning of an AI-based system for evaluating and improving the quality of textual requirements. The system classifies requirements as either SMART or non-SMART based on established guidelines, such as the INCOSE Requirements Working Group rules. Using Natural Language Processing (NLP) techniques and a BERT based model, the system recognizes NON-SMART requirements and suggests improvements towards making the requirements more Specific, Measurable, Achievable, Relevant, and Time-bound (SMART). The AI workflow includes several stages: first, the requirements are preprocessed and tokenized using a BERT tokenizer. The model is then fine-tuned on a labeled dataset of SMART and NON-SMART requirements using binary classification. To ensure robustness and avoid overfitting, cross-validation is used, along with model evaluation through performance metrics like accuracy, precision, recall, and F1-score. Additionally, a rule-based system is integrated to detect violations of specific INCOSE rules, and specific suggestions are provided for converting NON-SMART requirements into SMART ones. Designed with modularity and integration in mind, the system supports interoperability with MBSE workflows by facilitating traceability, consistency checking, and the generation of structured artifacts from unstructured inputs. This capability enables a more seamless transition from textual requirements to formal models within MBSE environments. The system’s performance is evaluated using confusion matrices and tested on an independent dataset to validate its performance and generalizability. Preliminary results of the AI-based system demonstrate good accuracy, showing immense potential for adoption towards improving requirements quality across various domains and seamlessly transitioning to MBSE workflows.

  • Towards a SMARTer Requirements: An AI-based Approach for Requirements Quality

    by Sharmila K.A., Ramakrishnan Raman

    It is well recognized that the quality of textual requirements plays a critical role in engineering of complex systems. When textual requirements suffer from issues such as ambiguity, inconsistency, and lack of verifiability, it causes significant issues in the systems engineering lifecycle of complex systems. This paper presents the design, development and fine-tuning of an AI-based system for evaluating and improving the quality of textual requirements. The system classifies requirements as either SMART or non-SMART based on established guidelines, such as the INCOSE Requirements Working Group rules. Using Natural Language Processing (NLP) techniques and a BERT based model, the system recognizes NON-SMART requirements and suggests improvements towards making the requirements more Specific, Measurable, Achievable, Relevant, and Time-bound (SMART). The AI workflow includes several stages: first, the requirements are preprocessed and tokenized using a BERT tokenizer. The model is then fine-tuned on a labeled dataset of SMART and NON-SMART requirements using binary classification. To ensure robustness and avoid overfitting, cross-validation is used, along with model evaluation through performance metrics like accuracy, precision, recall, and F1-score. Additionally, a rule-based system is integrated to detect violations of specific INCOSE rules, and specific suggestions are provided for converting NON-SMART requirements into SMART ones. Designed with modularity and integration in mind, the system supports interoperability with MBSE workflows by facilitating traceability, consistency checking, and the generation of structured artifacts from unstructured inputs. This capability enables a more seamless transition from textual requirements to formal models within MBSE environments. The system’s performance is evaluated using confusion matrices and tested on an independent dataset to validate its performance and generalizability. Preliminary results of the AI-based system demonstrate good accuracy, showing immense potential for adoption towards improving requirements quality across various domains and seamlessly transitioning to MBSE workflows.