Inspiration
India is home to 1.4B+ people, speaking 20+ official languages and hundreds of dialects. Yet critical information—on healthcare, agriculture, climate alerts—is often inaccessible or delayed. We were inspired to build a tool that breaks these barriers by creating AI-powered, multilingual videos in minutes, making knowledge fast, inclusive, and actionable.
What it does
SaarthiX is an AI-driven, multi-agent video generation platform that:
- Transforms text or raw data into mobile-first explainer videos.
- Supports multiple Indian languages and dialects.
- Auto-generates scripts, visuals, and voiceovers.
- Delivers ready-to-share content for social media, TV, and education platforms.
How we built it
- Designed a multi-agent system:
- Content Agent: Validates and curates data.
- Script Agent: Converts data into simple, local-language scripts.
- Video Agent: Generates visuals, captions, and voiceovers.
- Content Agent: Validates and curates data.
- Built an orchestration layer for seamless workflow between agents.
- Used FFmpeg and cloud rendering to deliver high-quality, lightweight videos.
- Optimized the platform for mobile-first and low-connectivity areas.
Challenges we ran into
- Handling real-time translation for Indian dialects.
- Managing GPU-heavy rendering pipelines efficiently.
- Ensuring cultural sensitivity in visual and audio content.
- Balancing speed, quality, and cost in production.
Accomplishments that we're proud of
- Built a fully functional AI video pipeline in under 48 hours.
- Delivered accurate multilingual scripts using open-source LLMs.
- Designed a no-code content upload system for NGOs and educators.
- Created a scalable architecture for nationwide use.
What we learned
- How to integrate multi-agent AI systems for creative workflows.
- Best practices for multilingual AI model training and evaluation.
- Techniques to optimize video generation for low-bandwidth environments.
- The importance of UX for diverse user bases, from farmers to city workers.
What's next for SaarthiX
- Expand to 20+ languages and add regional sign language support.
- Build an offline mode for rural areas with poor internet.
- Partner with government agencies, NGOs, and edtech startups.
- Enable real-time AI-driven video generation for emergencies (e.g., floods, heatwaves).
Built With
- Languages: Python, JavaScript, TypeScript
- Frameworks: React, FastAPI, Node.js
- AI Models: GPT-4o, Whisper, Stable Diffusion/Runway Gen-3, OpenAI TTS
- Cloud Services: AWS S3, GCP Video Intelligence API, Firebase
- Databases: MongoDB, PostgreSQL
- Tools: LangChain Agents, CrewAI, FFmpeg, Docker, Hugging Face
Built With
- crewai
- docker
- fastapi
- ffmpeg
- firebase-hosting-cloud-services:-aws-s3
- gcp-video-intelligence-api-databases:-mongodb
- github
- google-colab
- hugging-face-api-tools:-langchain-agents
- javascript
- languages:-python
- node.js-ai-models:-gpt-4o
- openai-tts-platforms:-hugging-face-hub
- postgresql-apis:-openai-api
- stable-diffusion/runway-gen-3
- typescript-frameworks:-react
- whisper
Log in or sign up for Devpost to join the conversation.