Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{197705,
        author = {Nikhil Pandey and Sonali Ramteke and Soham More and Sarvesh Mondkar and Atharva Nagthane},
        title = {MindEcho: An AI-Powered Multi-Modal Mental Health Support System with Real-Time Emotion Detection, Cognitive Distortion Analysis, and Crisis Risk Monitoring},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {7132-7137},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=197705},
        abstract = {Mental health remains one of the most under-served health challenges globally. Nearly one billion people live with a mental or neurological condition, yet most never receive professional care because of clinician shortages, geographic barriers, and social stigma. This paper introduces MindEcho, a full-stack AI mental health companion we developed to help bridge this gap. Every user message moves through a six-stage pipeline: (P1) multi-modal input handling for text, voice, and real-time live audio; (P2) emotion classification via the Distil Roberta transformer across seven emotion classes; (P3) crisis risk scoring through a weighted keyword engine with four severity levels; (P4) cognitive distortion detection grounded in Beck's CBT framework across ten distortion types; (P5) empathetic response generation using Llama-3-8B-Instruct via the Groq Cloud API; and (P6) multi-lingual text-to-speech output. Live sessions use Deepgram Nova-2 over WebSocket for sub-300 ms transcription. PHQ-9 and GAD-7 instruments are embedded for clinical grounding, and a mood heatmap calendar with per-session analytics enables longitudinal tracking. The platform runs on Render with a seven-table PostgreSQL database. Testing across 45 interactions recorded mean latency of 1.18 s for text and 1.48 s for live sessions, with average emotion detection confidence of 0.76.},
        keywords = {mental health AI; emotion detection; DistilRoBERTa; cognitive distortion; crisis detection; Groq; PHQ-9; GAD-7; Deepgram Nova-2; FastAPI; real-time speech recognition; CBT.},
        month = {April},
        }