🎬 BuzzBit-Mind AI-Powered Financial Psychology Platform

Automated short-form video generation about behavioral finance, investing psychology, and cognitive biases.

📱 What is BuzzBit-Mind?

BuzzBit-Mind is an end-to-end automated video generation pipeline that creates educational short-form content about financial psychology and behavioral finance. Every day, the system produces videos for YouTube Shorts, TikTok, and YouTube Long-form — in 6 languages (French, English, Spanish, Arabic, Hindi, Chinese).

🚀 Production Ready — Already generating 5+ videos/week

Daily automated content for YouTube Shorts, TikTok & YouTube Long-form in 6 languages.

🔧 Full Pipeline Architecture

📰 1. News & Theme Selection

Web scraping + Gemini Grounding → extracts evergreen psychological hooks from current events.

NewsAPI Gemini 2.5 Flash Web Search

🧠 2. LLM Script Generation

Mistral Small 4 / GPT-5.4 mini / Gemini → generates 9-10 segments scripts with strict word count (13-16 words/segment) and empathy injection.

Mistral Small 4 GPT-5.4-mini Chain of Thought

🎙️ 3. AI Voiceover

Voxtral Mistral (Studio quality, 44.1kHz) → natural voices in 6 languages.

Voxtral Mistral Goggle studio 44.1kHz Mono

🎬 4. Dynamic B-roll

Pexels/Pixabay integration → automatic video clips matching visual prompts.

Pexels API Semantic Search

📝 5. Subtitles

Whisper + forced alignment → word-level timing, multi-language subtitles.

Faster-Whisper FFmpeg

🎨 6. Thumbnails

Stable Diffusion + Mistral selection → 4 variants generated, human validation via Telegram bot.

Stable Diffusion LLM Selection

🤖 7. Telegram Bot

Human-in-the-loop validation → approves thumbnails, schedules publications, monitors comments.

Telegraf Webhooks

📤 8. TikTok API

Uploads videos directly to TikTok drafts. No manual intervention.

TikTok API OAuth 2.0

🔄 Step-by-Step Automation Flow

1
News Ingestion — Daily news fetch → Gemini extracts evergreen psychological concepts.
2
Theme Selection — Random pick from 210+ behavioral finance themes (Foundation, Momentum, Mastery levels).
3
Script Generation — LLM writes script respecting:
• 9-10 segments (Foundation=9, Momentum/Mastery=10)
• 13-16 words per segment
• Empathy phrase injected before CTA
• Logic connectors (unique per segment)
• JSON output format
4
Audio Generation — Voxtral (Mistral) or Google tts produces voiceover at 44.1kHz + Whisper creates word-level timestamps.
5
B-roll Assembly — Semantic search for clips matching visual prompts → downloaded and concatenated.
6
Video Rendering — Remotion compositing (audio + b-roll + subtitles + thumbnail).
7
Human Validation — Telegram bot sends 4 thumbnails for approval.
8
Publication — Video uploaded to TikTok API (draft) + YouTube (private).

🛠️ Technical Stack

Python 3.12 Node.js 20+ Remotion Mistral AI Mistral Voxtral OpenAI GPT-5.4 mini Google Gemini 2.5 Google TTS Faster-Whisper Pexels API Pixabay API Telegram Bot API TikTok API YouTube API FFmpeg Stable Diffusion LangChain

🎯 What We Need from TikTok API

✅ Video Upload

Direct upload to TikTok drafts using /v2/video/upload/ endpoint.

✅ OAuth 2.0 Authentication

Secure user authorization via https://www.tiktok.com/auth/authorize/.

✅ Video Metadata

Title, description, hashtags, privacy settings (draft/private).

❓ Why We Need TikTok API Access

🚀 Scale to 10+ videos/week

Manual upload is impossible at scale. API automation is the only path to daily content.

📊 Analytics Integration

We want to track engagement per video to improve our AI models.

⏰ Scheduled Publishing

Upload to drafts, then schedule at optimal times for each region.

🎬 Sample Video Output

Each video includes:

📹 Example video (French) — "Le biais d'attention sélective en finance"

YouTube Shorts (58s) | TikTok (90s) | YouTube Long (4min)

🔗 Resources

📧 Developer Information

👨‍💻 VolthaKiss — Full-stack AI Engineer
📧 volthakiss@gmail.com
🐙 github.com/volthakiss

© 2026 BuzzBit-Mind — Financial Psychology Education • Automated • Multilingual • Scalable

Built with ❤️ for TikTok Developer Challenge