Zora AI Models

The Deep Learning Engine Behind Our Products

Every Zora DeepTech product is powered by purpose-built, fine-tuned AI models — trained on domain-specific healthcare data, optimized for Indian clinical contexts, and deployed on GPU-accelerated infrastructure.

Two Products. Four Specialized Models.

Our AI stack is architected for healthcare precision

WellNest

Mental Wellness AI

ZoraWell-LLM

Gemma 3 serverless on Vertex AI

Live

ZoraEmote

Emotion + Voice AI

Phase 2

ZoraScan

Diagnostic Imaging AI

ZoraScan-Vision

CNN Anomaly Detection

Phase 2

ZoraReport-LLM

Gemma 3 serverless on Vertex AI

Phase 2
Model Cards

Detailed Model Specifications

ZoraWell-LLM

Mental Wellness Language Model for WellNest

Phase 1 Complete

Base Model

Gemma 3 (Google DeepMind)

Deployment

Serverless on Google Vertex AI

Context Window

128K tokens

Languages

English, Tamil, Hindi (Phase 2)

Task

Empathetic mental wellness conversation, crisis detection, personalized student support in Indian English + regional context

Phase 2 Roadmap

Fine-tune Gemma 4 on self-hosted NVIDIA GPU infrastructure for low-latency, offline-capable deployment

ZoraEmote

Voice & Emotion Detection Model for WellNest

Phase 2 Development

Architecture

Wav2Vec2 + LSTM Classifier

Inference

NVIDIA GPU (CUDA) — target <200ms latency

Task

Real-time emotion classification from student voice input (7 emotion classes: neutral, sad, anxious, stressed, happy, angry, fear)

Phase 2 Roadmap

Add real-time streaming inference via WebSocket for live session support

ZoraScan-Vision

Medical Imaging Anomaly Detection for ZoraScan

Phase 2 Development

Architecture

EfficientNet-B4 + YOLOv8

Inference

NVIDIA GPU (CUDA + TensorRT) — under 10 seconds per scan

Accuracy

94.7% sensitivity on NIH ChestX-ray14 benchmark dataset

Explainability

Grad-CAM heatmap overlay on every prediction

Task

Anomaly detection and localization in chest X-rays, brain MRIs, CT scans

Phase 2 Roadmap

Fine-tune with Indian patient scan data via hospital partnerships

ZoraReport-LLM

Clinical Report Generation Model for ZoraScan

Phase 2 Development

Base Model

Gemma 3 (Google DeepMind)

Deployment

Serverless on Google Vertex AI (Phase 1)

Task

Auto-generates structured clinical radiology reports from ZoraScan-Vision predictions and heatmap data

Phase 2 Roadmap

Self-hosted fine-tuned Gemma 4 on NVIDIA GPU — enabling offline report generation for hospitals with poor internet connectivity

Our AI Model Roadmap

Building the future of healthcare AI, one model at a time

Phase 1

Done

Completed

ZoraWell-LLM

Gemma 3 serverless on Vertex AI

WellNest MVP

Student mental wellness platform

Cloud Infrastructure

Google Vertex AI + Azure Central India

Phase 2

Active

In Progress (2025-2026)

ZoraEmote

Wav2Vec2 + LSTM on GPU

ZoraScan-Vision

EfficientNet + YOLOv8

ZoraReport-LLM

Gemma 3 on Vertex AI

ZoraWell-LLM v2

Fine-tune Gemma 4, Self-hosted NVIDIA GPU

Phase 3

Future (2027+)

ZoraIndic-LLM

Multilingual health AI — Tamil, Telugu, Hindi, Bengali

ZoraDrug-AI

Ayurvedic compound validation model (GNN)

On-premise Edge

Offline hospital deployment

How We Deploy Our Models

From cloud to edge, we build infrastructure for every scenario

Phase 1 — Cloud

Completed

Google Vertex AI

  • Gemma 3 serverless endpoints
  • AutoML pipelines
  • Vertex AI datasets

Azure Central India

  • Low-latency for Indian users
  • Data residency compliance
  • Backend services

Phase 2 — Hybrid

In Progress

Self-hosted NVIDIA GPU

  • Gemma 4 fine-tuned (self-hosted)
  • TensorRT optimized
  • CUDA inference layer

On-premise Hospital Edge

  • Offline ZoraScan
  • No internet needed
  • HIPAA-grade secure

Want to Integrate Our Models?

We offer API access for hospitals, diagnostic labs, universities, and healthcare platforms.