Sidharth
Kaliappan

Developing a heuristic evaluation framework to help developers recognize and mitigate data quality issues (completeness, accuracy, validity, contextuality) earlier in the development cycle. Grounded in a synthesis of 85 papers and refined through semi-structured developer interviews.

An LLM-powered conversational agent (Rasa framework, Claude 3.7 Sonnet) that guides users through designing structured, personalized, and safe self-experiments across health domains. Our within-subjects mixed-methods study with 42 participants showed a 19.2% improvement in experiment quality through theory-driven prompt engineering.

Extracted key caregiver concerns and support needs from 100,000+ posts in Alzheimer’s online communities using topic modeling (LDA). Engineered ML classifiers (SVM, Random Forest, Neural Networks) achieving AUC scores of 0.83-0.87 for detecting emotional and informational support patterns.

Conducted a comprehensive literature review identifying unmet needs of older adults in home settings. Collaborated with an interdisciplinary team to synthesize findings into evidence-based design recommendations for aging-in-place technologies.

Implemented a joint-learning framework using LayoutLMv1 for OCR and key-value extraction from 15,000+ printed and handwritten medical documents at Narayana Hrudayalaya Hospitals. Achieved Acc: 0.983, F1: 0.945 on real-world healthcare forms, enabling digitization of patient records.

Developed GAN and Wasserstein Autoencoder (WAE) models for synthetic pneumonia X-ray generation. Implemented a guided subset selection methodology to boost data diversity, improving classification accuracy by up to 21.64% on label-scarce datasets.

Designed a comprehensive player profiling pipeline using physiological signals (ECG, EDA, Respiration) and in-game events from 100+ Counter-Strike participants. Applied dimensionality reduction and model-based clustering (LDA, logistic regression) to identify gameplay-driven emotional response patterns.