# DIDI Backend

Misinformation detection backend — TypeScript/Node.js microservices.

## Services

| Service | Description | Port | Container |
|---------|-------------|------|-----------|
| **Agent V3** | Analysis engine — 4 detection components + verdict | 24803 | didi-agent-v3 |
| **didiFramework** | Parameters CRUD API, user management, Redis sync | 3005 | didi-framework |
| **Admin Dashboard** | React admin UI (services, config, history, users) | 3000 | didi-admin |
| **Kong** | API Gateway (SSL, routing, rate limiting) | 443 | kong |
| **Keycloak** | Authentication (OAuth2/OIDC, JWT) | 28000 | keycloak |
| **Redis** | Cache (framework config, session state) | 6379 | didi-cache |
| **RabbitMQ** | Queue (async analysis, 24 component queues) | 5672 | rabbitmq |
| **MinIO** | Object storage (media files, per-user buckets) | 9000 | minio |
| **PostgreSQL** | External cluster (Patroni/HAProxy) | 5000 | — |

## Quick Start

```bash
# 1. Start data layer
cd services/data-layer && docker compose up -d

# 2. Start agent-v3 + workers
cd services/orchestration-layer/agent-v3 && docker compose up -d --build

# 3. Start framework
cd services/orchestration-layer/didiFramework && docker compose up -d --build

# 4. Sync parameters
docker exec didi-framework sh -c 'wget -qO- --post-data="" http://127.0.0.1:3005/api/sync-redis'

# 5. Test
curl http://localhost:24803/api/v3/health
```

## Structure

```
backend/
├── admin-dashboard/          # React 19, MUI 7 — admin UI
├── production/               # Production docker-compose + .env
└── services/
    ├── data-layer/           # Redis, RabbitMQ, MinIO, local PG
    ├── gateway-auth-layer/   # Kong + Keycloak
    └── orchestration-layer/
        ├── agent-v3/         # Analysis engine (TypeScript, Express 5)
        └── didiFramework/    # Parameters API (TypeScript, Express 4)
```

Each service has its own `INDEX.md` with detailed documentation.

## Analysis Pipeline

```
Input (text/image/audio/video/URL)
  │
  ├─ Worker: media-preprocess (extract text from media)
  │
  ├─ Worker: techniques     → manipulation_score (0-100)
  ├─ Worker: ai-tampered    → ai_probability (0-100)
  ├─ Worker: claims         → credibility_score (0-100)
  ├─ Worker: domain         → trust_score (0-100)
  │
  └─ Aggregator: verdict    → risk_score (0-100) + explanation (RO+EN)
```

All analysis endpoints are **async**: dispatch to RabbitMQ → return 202 → poll for results.

## Rebuild Commands

```bash
# Agent V3
cd services/orchestration-layer/agent-v3
docker compose up -d --build agent-v3

# Framework
cd services/orchestration-layer/didiFramework
docker compose up -d --build didi-framework

# Admin Dashboard
cd admin-dashboard
docker build -t didi-admin:latest .
cd ../services/data-layer && docker compose up -d --force-recreate didi-admin
```

## Credentials (Staging)

| Service | User | Password |
|---------|------|----------|
| PostgreSQL (cluster) | bos_interface | interface |
| Redis | — | redis123 |
| RabbitMQ | admin | rabbitmq123 |
| MinIO | minioadmin | minio123 |
| Keycloak | admin@didi.local | admin123 |