AI-Brand-Protect

An AI-powered brand onboarding agent for brand protection platforms.

Given only a company name and one or more official domains, it automatically discovers and compiles a comprehensive BrandProfile.

Tech Stack

Layer	Technology
Runtime	Python 3.13, asyncio
Schemas	Pydantic
LLM	OpenAI
Search	Tavily API
Domain verification	IANA RDAP bootstrap
API server	FastAPI + Uvicorn
Containerisation	Docker

Pipeline Flow

brand_input.json  (or POST /scan body)
│
main.py / server.py
│  (validate inputs)
│
asyncio.gather - five extractors run concurrently:
├── extract_domains -> Tavily search -> LLM -> RDAP validation
├── extract_assets -> HTTP scrape (og:image, favicons)
├── extract_keywords -> Tavily search -> LLM (GICS taxonomy)
├── extract_people -> Tavily search -> LLM structured extraction
└── extract_socials -> Tavily search -> LLM URL extraction
│
BrandProfile (Pydantic)
│
├── API mode  -> JSON response body
└── CLI mode  -> output/<company>_brand_profile.json

Output Schema

Field	Description
discovered_domains	Related domains with RDAP ownership confirmation
visual_assets	Logo and favicon URLs with source and type
keywords	GICS-mapped industry taxonomy terms
key_people	Executives with name, role, and source URL
social_links	Official social media profile URLs

Potential Next Steps

1. Vision Model: Pass discovered logo URLs to a vision model to confirm logo for [Company].
2. DNS Validation: Resolve discovered domains against public DNS to filter parked or stale domains.
3. Agentic Loop: Replace single-pass extraction with an agentic loop for refined & validated results.
4. Log Redaction: Custom logging formatter that scrubs API key patterns before output.
5. LLM Fallback: Support a secondary model if the primary is unavailable.
6. Contact Emails: Extend social extraction to also get official customer support emails.

See how_to_run.md to get started. See analysis.md for implementation decisions.