Integration
How NSFA AI works, and how platforms, NGOs, and researchers can use it
NSFA AI is a multimodal content moderation system. Rather than relying on a single signal, it combines computer vision, natural language processing, and large language model reasoning to assess whether a piece of social media content contributes to wildlife exploitation.
Every analysis produces a calibrated risk score (0–1), an interpretable classification label, and 30+ structured fields — from detected species and environment to trafficking language, evasion tactics, and facility ethics. This structured output is what makes the system integrable: platforms can route content, NGOs can build intelligence, and researchers can query a consistent dataset.
The system operates as a 4-stage pipeline:
Input
Image upload or a social media URL (TikTok, Instagram, YouTube, Facebook, X). Optional caption, profile description, and location metadata.
Vision Module
Species identification, environment classification (wild, zoo, tourism, domestic…), human presence, and interaction type analysis.
Context Module
Caption NLP for code words, emoji substitution, transaction intent, and private-channel redirects. Poster credibility and facility ethics lookup.
LLM Reasoning
Multimodal chain-of-thought synthesis producing a composite risk score and a structured JSON record with 30+ fields.
Each analysis returns a structured record. Key fields:
// Analysis record
{
"risk_score": 0.87,
"label": "likely_unethical",
"species_detected": ["tiger"],
"environment": "tourism",
"human_present": true,
"interaction_type": "posing / physical contact",
"trafficking_language_detected": false,
"transaction_intent_explicit": false,
"evasion_tactics_detected": true,
"wildlife_products_detected": false,
"facility_ethics_assessment": "unethical",
"ai_reasoning": "...",
"moderation_status": "pending"
}
The full field reference is documented in the white paper (Tables 1–4, Section 4.2).
Platform Moderation
Route user-generated content through NSFA AI at upload to pre-screen wildlife content before publication or during appeals.
Conservation Monitoring
NGOs can surveil hashtags and accounts associated with suspected trafficking or tourism exploitation.
Research & Datasets
Researchers receive structured, labelled records for downstream work in animal-welfare AI and multimodal ethics.
Enforcement Intelligence
Aggregated, anonymised trend data supports agencies enforcing CITES and cross-border wildlife crime.
NSFA AI is a hosted proof-of-concept operated as an independent initiative — not a public, self-hostable codebase. The complete methodology, risk scoring rubric, and output schema are published in the white paper. Partner organisations can request access to pilot the live system and integrate its outputs.
Interested in integrating NSFA AI? Read the white paper