EchoSync

Ambient Intelligence & Real-Time Proximity Matching Platform

EchoSync is a privacy-first, low-latency proximity discovery system that passively listens to ambient conversations, sanitizes personal identifiable information (PII) on-device, extracts vector interest tokens, and matches nearby users in real-time within a 15-meter radius.

---

Key Features

• On-Device PII Scrubbing: Utilizes local LLMs (Llama-3-8B) to redact sensitive data before any vector embedding is processed.
• 15-Meter Match Radar: Real-time spatial tracking and visual sonar interface displaying passive and active connections.
• Ephemeral Match Feed: Self-destructing match cards governed by a strict 15-minute Time-To-Live (TTL).
• AI-Generated Icebreakers: Contextual, dynamic prompt suggestions designed to initiate immediate, real-world conversations.
• Haptic Proximity Feedback: Subtle physical cues sent to mobile clients upon detecting high-similarity interest vectors.

---

Architecture & Technology Stack

Frontend & Dashboard

• Mobile App: Built with Flutter (Cross-platform iOS/Android) featuring native haptic engine integration and custom radial canvas drawing.
• Command Center: Web dashboard built with React / Next.js, Tailwind CSS, and Lucide Icons for operational monitoring.

Backend & AI Pipeline

• Core API: High-performance REST & WebSocket server built with Go (Golang).
• Privacy Firewall: Local Llama-3-8B inference engine running on-device or via edge worker node for zero-trust transcript scrubbing.
• Proximity Engine: Geo-fencing & spatial vector indexing using Redis Geo and pgvector (PostgreSQL).

---

Wireframe & UI Overview

Desktop Command Center	Mobile Radar & Icebreaker View
Multi-panel operations dashboard featuring live PII pipeline, spatial radar, and ephemeral feed.	Flutter mobile layout showing real-time proximity radar and generated icebreaker prompts.

---

Quick Start & Installation

Prerequisites

• Go 1.22+
• Flutter SDK 3.19+
• Node.js 20+ & pnpm (npm works too)
• Docker & Docker Compose

1. Environment configuration

cp .env.example .env

The Go server reads its configuration from environment variables, so export them before running the backend:

set -a; . ./.env; set +a

Ollama (Llama-3-8B) is optional. When unreachable, the server automatically falls back to the deterministic rule-based PII firewall and template icebreakers.

2. Static analysis

flutter pub get
flutter analyze

3. Start the infrastructure

docker compose up -d postgres redis   # postgres :5544, redis :5545

4. Start the backend

set -a; . ./.env; set +a
cd backend && go run ./cmd/server     # listens on :8080

Verify it is up:

curl http://127.0.0.1:8080/health

5. Start the command center (Next.js / React)

cd web
npm install
npm run build
npm start                             # http://localhost:3000

The command center talks to the backend at 127.0.0.1:8080 by default. Override it with NEXT_PUBLIC_API_URL before running npm run build.

6. Start the Flutter app

One-time web platform setup (adds Flutter web files next to the Next.js command center — do not run flutter create . again after this):

flutter create . --platforms web

Build and serve:

flutter build web --release
python3 -m http.server 3001 --directory build/web   # http://localhost:3001

Or run on a connected device/emulator with flutter run.

Live mic capture: the mic button streams pcm_s16le @ 16 kHz mono to the backend's WS /v1/audio/stream and shows live captions. Mic streaming works on Android/iOS/desktop builds; on Flutter web (browser cannot emit raw PCM) the button shows a hint and the radar demo still runs.

7. End-to-end demo

bash scripts/demo.sh   # requires the backend on :8080 and jq

7b. Audio pipeline hands-on (WAV upload)

# Transcribe an uploaded clip -> scrub PII -> register presence -> match:
curl -XPOST 'http://127.0.0.1:8080/v1/audio?user_id=kim&lat=-1.282&lng=36.821' \
  -H 'Content-Type: audio/wav' --data-binary @/path/to/speech.wav

• One-shot uploads are transcribed with Groq Whisper (whisper-large-v3-turbo, falls back to a Speechmatics short session for raw PCM).
• Live streaming (mic, WS) runs through Speechmatics real-time (wss://eu.rt.speechmatics.com/v2) with final transcripts debounced and fed into the same scrub -> presence -> match engine.
• GROQ_MODEL selects the cloud Llama-class firewall (default openai/gpt-oss-120b); ECHO_PII_MODE=auto prefers it, then Ollama/rule.
• Privacy pipeline events stream over /ws?user_id=dashboard and /v1/pipeline/events for the command center.

Service summary

Service	URL / Port
Backend API	http://127.0.0.1:8080
Next.js command center	http://127.0.0.1:3000
Flutter app (web)	http://127.0.0.1:3001
PostgreSQL	localhost:5544
Redis	localhost:5545