MConAI: Multilingual Conversational AI for Low-Cost Early Alzheimer's Detection

Chennareddy, Srihith

doi:10.1145/3789982.3790060

About

I build AI systems for accessible healthcare - from multimodal models for early Alzheimer’s detection to conversational AI for cognitive screening. As a Simons Fellow '26 and first author researcher with publications in IEEE and ACM, I work at the intersection of machine learning, neuroscience, and healthcare AI. My work has been recognized at MIT Lincoln Labs, presented at international conferences, awarded a Samsung research grant, and featured in the AAIC Neuroscience Next program.

Research & Publications

Keck School of Medicine of USC • 2025 — 2026

MConAI: Multilingual Conversational AI for Low-Cost Early Alzheimer's Detection

Architected a multilingual conversational AI framework for early Alzheimer’s detection using mBERT and conversational transcripts from DementiaBank, achieving an average accuracy of 87% in English and 83% across all languages with multiclass classification.

First Author ACM ACM Digital Library Video

Foundation AI Models for Reliable MRI-Based Dementia Detection

Developed foundation vision transformer models (ViT, 3D-Swin) for early Alzheimer’s detection from structural MRI scans, achieving 91.5% accuracy.

First Author IEEE IEEE Xplore Video

Affordable Conversational AI for Global Dementia Screening

Presented scalable, low-cost conversational AI approaches for dementia screening for global populations at the Alzheimer's Association International Conference (AAIC) Neuroscience Next 2026.

First Author AAIC AAIC Neuroscience Next Video

Attention-Augmented Temporal Fusion Transformer for Short-Horizon Hypoglycemia Prediction in Type 1 Diabetes

An attention-based Temporal Fusion Transformer predicts 30-minute hypoglycemia in Type 1 Diabetes with 91% accuracy, outperforming traditional and glucose-only models. Enables reliable real-time prediction for safer, proactive glucose management.

AACE ScienceDirect AACE Endocrine Practice

University of Washington, Seattle • 2025 — 2026

CropViT: A Climate-Aware Vision Transformer Model for Crop Yield Prediction

Developing affordable computer vision models integrating climate data with vision transformers for predictive crop yield estimation and disease detection in precision agriculture.

First Author IEEE IEEE Xplore Video

Optimizing AI Model Training Costs with Stratified Sampling and Self-Adaptive Testing

Investigating cost-efficient approaches to AI model training through stratified sampling techniques and self-adaptive testing frameworks to reduce computational overhead without sacrificing accuracy.

First Author IEEE IEEE Xplore Video

MIT Lincoln Labs • 2025 — 2026

FOCAL: Adaptive Framework for Cognitive ADHD Learning

Engineered FOCAL, an adaptive AI attention assistant using multimodal wearable signals to optimize task initiation and manage cognitive load for neurodiverse individuals.

Finalist Final Event 2026 Video