CreatorScout

Finds and scores regional gaming micro-influencers across YouTube, TikTok, and Instagram, enabling small teams to build targeted outreach lists in minutes instead of days.

---

Screenshots

---

The Problem

Partnering with regional micro-creators in local European markets (e.g., Finland, Germany) presents two major challenges:

• Language Barriers: Local creators title videos in their native language (halpa pelikone, günstiger gaming pc), rendering English searches ineffective.
• Manual Overhead: Vetting individual creator metrics, engagement, and content alignment by hand does not scale.

---

What It Does

1. Multi-Language Search: Translates a search brief (e.g., "budget PC build") into 8+ European languages automatically.
2. Multi-Platform Scraping: Collects public creator data across YouTube, TikTok, and Instagram.
3. Rolling Averages: Calculates 30-day and 90-day average view counts to filter out creators inflated by single viral hits.
4. AI-Powered Fit Scoring: Analyzes video transcripts, content niches, and risk factors using LLMs to score creators from 0–100.
5. CSV Export: Generates an actionable CSV containing metrics, bio/contact emails, country data, and AI fit scores.

---

Workflow Architecture

Search Brief
   │
   ▼
Translate Brief (8+ Languages)
   │
   ▼
Data Collection (YouTube / TikTok / Instagram)
   │
   ▼
Data Cleaning & Windowed Averages (30d / 90d Views)
   │
   ▼
LLM Scoring & Risk Analysis (With Model Fallbacks)
   │
   ▼
Streamlit Dashboard ──► Export CSV

Key Technical Decisions

• LLM Model Fallback System: Requests target a primary Groq model and automatically cascade to secondary models during rate limits or API outages.
• Windowed Metrics over Total Counts: Evaluates sustained viewer engagement rather than vanity totals to accurately assess audience quality.

---

Tech Stack

Tool / Library	Usage
Streamlit	Web application dashboard and cloud deployment
ytscrape	Fetches YouTube channel metadata, view counts, and transcripts
requests + BeautifulSoup	Scrapes public TikTok metadata
ddgs	Discovers public Instagram profiles and Reels via DuckDuckGo Search
pandas	Data manipulation, rolling 30/90-day view metrics, and CSV generation
Groq API	High-speed LLM inference for fit scoring and transcript analysis
Cursor / Claude / Gemini	AI development tools utilized during development

---

Local Setup

1. Clone the repository

git clone https://github.com/Samir-BK/Prompt-Hackathon.git
cd Prompt-Hackathon

2. Set up a virtual environment

python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Configure environment variables

Create a .env file in the project root:

echo 'GROQ_API_KEY="your_groq_api_key_here"' > .env

5. Launch the application

streamlit run app.py

---

Current Limitations

• Web Scraping Dependency: Relies on public HTML structures and search indexes, which are vulnerable to layout updates. Official APIs should be integrated for production use.
• LLM Ranking Nature: AI scores serve as an automated preliminary filter; manual review of shortlisted candidates remains recommended.
• Rate Limits: Rapid sequential searches may hit rate limits on search engine scraping modules.

---

Roadmap

• Direct email outreach integration via Gmail / Outlook API
• Conversion tracking pipeline to refine LLM scoring prompts over time
• Extended support for Twitch and X (Twitter) profile discovery