Koustuv Saha

Across events, disruption onset was associated with immediate increases in anxiety, stress, suicidal expression, and grief activation, with relational discontinuity and transition-support deficit being associated with more adverse immediate responses across several outcomes.

CareMirror is built, an envisioned caregiver wellbeing ecosystem with interconnected caregiver- and clinician-facing interfaces for longitudinal reflection, personalized support, and caregiver-controlled sharing with clinical care.

A co-design study where 14 participants redesigned Google and Apple Health interfaces using Figma Make, finding conversational AI designs converged around chat-window conventions and implications for preserving agency and flexible user-directed interfaces.

PeerPen, a writing assistance tool embedded within a Reddit-like interface, supporting two main features: draft generation and revision of user-written responses, is built and it is found that PeerPen reduced the burden of composing responses and increased confidence in offering support.

Automated feature engineering for heart failure can boost predictive accuracy by over 7% while maintaining rigorous evidence standards.

LLMs morph into riskier conversationalists when playing directive support roles like "Coach" or "Inform" for caregivers, revealing a troubling quality-safety trade-off.

Open-source LLMs, trained with Reddit community feedback, can match the performance of proprietary models in generating supportive mental health responses, offering a privacy-preserving alternative.

Content moderation systems, designed to protect users, may inadvertently cripple LLMs' ability to function as effective therapists by censoring crucial topics in real therapy sessions.

AI negotiation coaches don't work for everyone: your personality determines whether you'll benefit from AI or a good old-fashioned handbook.

LLMs used in matchmaking amplify existing caste hierarchies, rating same-caste matches significantly higher and perpetuating social biases in sensitive relational domains.

AI-agent communities aren't just pale imitations of human ones; they're structurally and linguistically distinct, exhibiting extreme inequality and homogenization driven by identifiable agent-level stylistic outliers.

Generative search engines create "answer bubbles" by selectively citing and framing information, leading to divergent information realities compared to traditional search.