Migoo

Migoo

📖 Inspiration

Pharma marketing teams spend ₹5-10L and 4-6 weeks creating patient education & HCP outreach videos.
90% get <10% engagement due to generic content and slow production.

Migoo solves this - raw communication briefs → production-ready MP4s in <5 mins using open-source AI.
Perfect for Indian pharma (Sun Pharma, Dr. Reddy's, Cipla) - Hindi/English compliant videos at 1/100th cost!

"Brief → Pro video in 5 mins. Pharma marketing, disrupted." 😎

🎥 What it does

Input: Pharma communication brief (Google Doc, Notion, raw text)
Output: 30-60 sec compliant MP4 with voiceover, animations, disclaimers "New diabetes drug mechanism for cardiologists" ↓ 45 secs later ↓ 🎬 Animated mechanism + HCP voiceover + stats overlayHindi subtitles + "Consult physician" disclaimer Core Features:

✅ Brief Parser - Extracts facts, audience, CTAs from messy docs
✅ Smart Scriptwriter - Scene-by-scene breakdown + visual prompts
✅ Text-to-Video - 720p clips w/ character consistency
✅ Pharma Voice - Professional Hindi/English TTS
✅ MLR Compliance - Auto-flags regulatory violations
✅ Multi-format - Reels/LinkedIn/Full video exports

🛠️ How we built it

100% Open Source Stack (No OpenAI/Gemini):

Layer	Tech	Role
Parse	Llama3.1 8B	Brief → structured JSON
Script	Qwen2-VL 7B	Scene breakdown + prompts
Video	Open-Sora 1.3	5-sec 720p clips @ 30fps
Audio	Piper TTS	Pharma-grade voiceovers
Edit	FFmpeg + MoviePy	Pro stitching + subtitles
UI	Streamlit + FastAPI	Drag-drop interface

Pipeline Flow:

graph LR
    A[📄 Brief Input] --> B[Llama3.1 Parser]
    B --> C[Qwen2 Script Gen]
    C --> D[Open-Sora Video]
    D --> E[Piper Audio]
    E --> F[FFmpeg Editor]
    F --> G[⚕️ Compliance Check]
    G --> H[🎥 MP4 Output]Deploy: Docker → Azure VM → Streamlit sharing

⚠️ Challenges we ran into
Video Flickering - Open-Sora inconsistent scenes
Fixed: Image-to-Video chaining (frame1 → frame2)Pharma Compliance - No MLR-trained models
Fixed: BERT fine-tuned on 500+ Indian pharma adsVoice Quality - Robotic TTS
Fixed: StyleTTS2 + custom "HCP voice" datasetSpeed - 15 mins/video initially
Fixed: ONNX + Azure GPU = 2.3 mins/videoHindi Accents - Poor regional voices
Fixed: IndicTTS on HCP consultation data