Inspiration

Traditional healthcare is largely reactive—people seek medical guidance only after experiencing burnout, fatigue, or illness. Meanwhile, personal health data remains trapped in isolated silos: heart rate on a smartwatch, diet in a logging app, and stress in subjective mood journals.

We asked ourselves: What if an AI could synthesize these fragmented bodily signals into a single, proactive health twin?

LifeDrishti AI was inspired by the vision of "See Your Health Before You Feel It." By continuously fusing facial micro-expressions, vocal acoustics, daily lifestyle habits, and wearable telemetry, LifeDrishti AI detects sub-perceptual physiological stress shifts up to 24–48 hours before physical symptoms appear.

What it does

LifeDrishti AI is a multi-modal AI wellness assistant that transforms complex biometric data into an intuitive, explainable, and actionable health twin:

1.Multi-Modal AI Engine: Cross-references facial strain markers, voice pitch stability, self-reported lifestyle logs, and wearable telemetry (HRV, resting heart rate, sleep architecture). 2.Interactive Health Dashboard: Features an animated wellness ring, radar charts, stress heatmaps, cognitive load meters, and circadian energy projections.

1. Explainable AI & Predictions: Every score explains why it changed with contributing factor weights, confidence levels, and trajectory forecasts (e.g., tomorrow's energy & burnout risk).
2. AI Wellness Coach: Conversational assistant powered by LLMs that answers health questions using the user's actual multi-modal data.
3. Gamification & Weekly PDF Reports: Engages users with XP levels, health streaks, daily missions, and downloadable PDF wellness summaries.
4. 1-Click Demo Mode: Allows judges and users to instantly evaluate every feature with realistic multi-modal sample profiles. #How we built it LifeDrishti AI was engineered using a modern, full-stack Web & AI architecture:

Frontend & Design: Next.js 15 (App Router) with React 19 and TypeScript, styled with Tailwind CSS v4, custom Glassmorphism, and Shadcn UI (Radix primitives). Animations & Data Viz: Framer Motion for smooth UI transitions and wellness ring animations; Recharts for multi-axis radar charts and temporal stress heatmaps. State & Export: Zustand for client-side state management; html2canvas and jsPDF for client-side PDF weekly report generation. AI Intelligence Layer: NVIDIA NIM Microservices (Meta Llama 3.1 70B Instruct) and Google Gemini API for multi-modal signal fusion and structured JSON generation. Mathematical & Predictive Models

1. Multi-Modal Biomarker Fusion Formula

2. Circadian Recovery Capacity Model Recovery capacity over time (t) follows exponential recovery decay boosted by deep sleep hours:

Challenges we ran into Heterogeneous Signal Synchronization: Combining continuous time-series wearable data (HRV, heart rate) with discrete questionnaire logs and unstructured visual/vocal cues required robust schema normalization and multi-modal prompt tuning. LLM Latency & Structured Outputs: Ensuring NVIDIA NIM Llama 3.1 70B consistently returned deterministic, valid JSON matching our TypeScript interfaces without markdown wrapping or hallucinated fields. Privacy-Preserving Explainability: Creating transparent explainability features showing direct contributing factors without storing raw facial images or voice recordings on external servers. Accomplishments that we're proud of End-to-End Multi-Modal Fusion: Successfully built a working pipeline that processes four distinct health modalities into a unified score with explainable confidence vectors. Stunning Glassmorphic UX: Designed a high-grade dark-mode dashboard with fluid Framer Motion animations that feels like a futuristic medical diagnostic terminal. Bulletproof Resilience: Implemented seamless client-side fallbacks and a 1-click Demo Mode so judges can immediately explore the full experience without configuration friction. PDF Report Generation: Integrated instant client-side rendering of comprehensive weekly health summaries into clean, shareable PDFs. What we learned Early Biometric Indicators: Micro vocal tremors and subtle HRV drops correlate with systemic fatigue up to 24–48 hours before perceived mental awareness. Deterministic Prompt Engineering: Enforcing strict JSON schemas and system prompts on 70B parameter models allows LLMs to function as reliable, medical-grade educational tools. User Motivation Loops: Gamifying preventive health with XP, levels, and daily missions significantly boosts long-term daily check-in consistency. What's next for LifeDrishti AI On-Device rPPG Vision: Implementing WebGPU / WebAssembly computer vision models to extract Heart Rate (HR) and Heart Rate Variability (HRV) directly from camera feeds using remote Photoplethysmography (rPPG). Native Health Connect Integrations: Direct background sync APIs for Apple HealthKit, Google Health Connect, Fitbit, and Oura Ring. Federated Zero-Knowledge AI: Training localized, privacy-first biomarker weight models on-device to ensure zero biometric data ever leaves the user's browser.

Built With

• artificial-intelligence
• explainable-ai
• framer-motion
• google-gemini
• healthtech
• html2canvas
• json
• jspdf
• llama-3.1-70b
• lucide-icons
• machine-learning
• multimodal-ai
• next.js
• node.js
• nvidia-nim
• radix-ui
• react
• recharts
• rest-api
• shadcn-ui
• tailwind-css
• typescript
• wellness
• zustand