Rishav Mallick

Project Title: Krishi Mantrana (AI Based Multimodal Dialog System for Farmers)

• Coordinated the development of a chatbot in Bengali and Hindi for a multimodal question-answering system.
• Sentence transformers were fine-tuned using a siamese network topology, and distance measures were employed to ascertain the similarity between two sentences.
• Built a data collection web application for model training purposes and deployed using Flask web application for the project and deployed it to production.
• Customized ASR and TTS systems for Indic languages.

Project Title: Odia Document OCR System Development

• Performed Image Pre-processing tasks using OpenCV for enhanced Odia text extraction.
• Adopted a state-of-the-art OCR model for accurate Odia text recognition, based on CNN-LSTM.
• Developed Web API for seamless image-to-text conversion for Odia language documents.
• Designed and developed GUI for intuitive image-to-text conversion for Odia language.
• Achieved 94% character-level accuracy through rigorous model testing over different datasets.

Project Title: ASMA App - Autism Screening and Awareness

• Developed ASMA App for early autism detection in children aged 2 to 12.
• Features: Awareness, early signs recognition, clinic locator.
• Supervised the data acquisition from special schools, NGOs, and regular schools.
• Build ML models including Naïve Bayes, KNN, Random Forest, and SVM for autism screening, with Random Forest achieving the highest accuracy of 97%.

Project Title: XOR-based Text Decryption Algorithm Development

• Over 50,000 domain-specific sentences were extracted by means of a rigorous web crawling procedure from publicly available sources, and data processing techniques were employed for the purpose of data refining.
• Designed and optimized a Random Forest model for character prediction, achieving an average accuracy of 86%.
• Conducted detailed analysis of Language Models, leveraging SRILM for algorithmic enhancements and developed innovative algorithm designs integrating SRILM Language Model, elevating text processing capabilities.
• Engineered an intuitive GUI to showcase algorithm performance on text streams, enhancing user interaction and comprehension.