Skip to content

Native AI

native is a mobile-first traveler intelligence agent built for people entering unfamiliar places in India.

Code Video Cerebral Valley

Core trick

Auto-detects incoming speech language and translates only when it differs from the user's home language, with a risk classifier on transcripts.

Tags

voicevisionsecurity

Description

native is a mobile-first traveler intelligence agent built for people entering unfamiliar places in India. The core problem it solves is the “arrival gap”: language barriers, local process confusion, and safety risk in real-time interactions. The app has 3 primary traveler modes: Listen, Vision, and Navigate. In Listen, it captures live speech, auto-detects incoming language, and translates/explains only when needed in the user’s selected home language. In Vision/Navigate, it interprets signs/surroundings and gives concise route + action guidance. It includes a Local Risk Guard that detects scam/coercion/urgency cues and gives localized next steps, with city-aware safety guidance and emergency action support (India metro-focused). A secondary tab, BAKSHI, captures oral agreement terms into structured points.
DivyamJindal/native-become-a-native-on-second-0★ 00 forksTypeScriptApache-2.0Created Feb 14, 2026 · last push Feb 14, 2026

README excerpt

# native

`native` is an India-first realtime multimodal app built for Gemini 3 Bengaluru Hackathon (Saturday, February 14, 2026).

The app helps people become locals faster in unfamiliar places by combining what they see, hear, and ask in one live experience.

## Hackathon provenance

- Built during event hours for Gemini 3 Bengaluru Hackathon.
- Based on the official Google Gemini Live web starter as a foundation.
- This repository contains only open-source code and assets.

## Features

- `Spot` mode: camera-first live reading and translation.
- `Echo` mode: ambient speech translation with translation-only output behavior.
- `Guide` mode: camera + voice + Google Search grounding with stepwise outputs.
- `Bridge` mode: bidirectional live interpretation (primary demo pair Hindi <-> Kannada).
- User toggles for every core feature: session, mic, vision input, search, and mode.
- Deterministic 3-scene demo presets with in-app script hints.
- Live transcript panel with input and output transcription.
- Reliability states for permission issues, reconnect flow, and retry action.

## Language coverage (current UI)

- Hindi (`hi-IN`)
- Kannada (`kn-IN`)
- English (`en-IN`)
- Bengali (`bn-IN`)
- Marathi (`mr-IN`)
- Tamil (`ta-IN`)
- Telugu (`te-IN`)

## Quick start

1. Create `.env` in project root:

```bash
REACT_APP_GEMINI_API_KEY=your_api_key

# Optional model override
# REACT_APP_GEMINI_MODEL=models/gemini-2.5-flash-native-audio-preview-12-2025

# Optional default language pair f

Team