Navigate IDD
Role: AI Systems Product Designer | Timeline: Sep 2025 - Oct 2025
Tech Stack: n8n, OpenAI GPT-4 & Claude Sonnet 4.5, Pinecone, Synthesia
Overview
Navigate IDD is a governed AI system designed to transform complex Medicaid waiver regulations into verified, multilingual educational videos. In a domain where accuracy is critical, this system bridges the gap between dense policy documents and accessible family guidance, ensuring compliance while drastically reducing content production time.
The Challenge
Families seeking disability services often face a barrier of dense regulations and confusing waiver language.
- Accessibility: Critical information is locked in complex legal text, often unavailable in languages other than English.
- Scalability: Traditional video production is slow, expensive, and difficult to update when policies change.
- Compliance: Generative AI solutions often struggle with "hallucination," which is unacceptable in regulated healthcare environments.
The Solution
I architected and built an end-to-end Retrieval-Augmented Generation (RAG) pipeline that grounds AI outputs strictly in verified policy documents. The system automates the creation of scripts, translations, and video production, all while maintaining a rigid human-in-the-loop governance model.
System Architecture
-
Policy Knowledge Layer: Official Medicaid waiver files are indexed in Pinecone, creating a searchable vector database that prevents hallucination.
-
RAG Script Generation: AI agents retrieve relevant policy sections to draft accurate, structured scripts.
-
Human Governance: A dedicated dashboard allows subject matter experts to review and approve scripts before production.
-
Multilingual Production: Approved scripts are translated and converted into video explainers using Synthesia, ensuring consistent tone and meaning across languages.
-
Automated Distribution: The system handles metadata generation and publishing to social platforms (YouTube, etc.).
Key Features
| Category | Feature | Description |
|---|---|---|
| Accessibility | Plain Language | 6th grade reading level for maximum comprehension. |
| Multilingual Support | English, Spanish, Chinese, Korean, Vietnamese. | |
| Caption Support | Automatic caption generation for hearing accessibility. | |
| Cultural Sensitivity | Adapted content for diverse communities. | |
| Accuracy & Governance | Human Approval Gates | Multiple review points for accuracy. |
| Source Verification | RAG system ensures information accuracy. | |
| Expert Review | Subject matter expert validation before publication. | |
| Feedback Loop | User comments routed back for content improvement. | |
| Scalability & Automation | Automated Workflow | Minimal manual intervention required. |
| Multi-platform Distribution | Simultaneous posting across social platforms. | |
| Template System | Reusable components for different content types. | |
| Monitoring Dashboard | Real-time status tracking and analytics. |
Technical Stack
| Category | Technology | Purpose |
|---|---|---|
| Core Technologies | n8n | Workflow orchestration and automation. |
| OpenAI GPT-4 & Claude Sonnet 4.5 | AI script generation and translation. | |
| Pinecone | Vector database for RAG implementation. | |
| Synthesia | AI video generation platform. | |
| Integrations | Google Workspace | Drive, Docs, Sheets for document management. |
| YouTube API | Video upload and metadata management. | |
| Social Media APIs | Multi-platform distribution. | |
| Webhook System | Real-time approval and status updates. |
Impact & Results
The system validated that AI can responsibly scale public sector education.
| Metric | Result |
|---|---|
| Content creation time | ~90% reduction |
| Script generation accuracy (policy grounded) | 100% source traceable to verified policy text |
| Languages supported | 5 (English, Spanish, Chinese, Korean, Vietnamese) |
| Average time from topic upload to script draft | Under 2 minutes |
| Average time from script approval to video file | Under 15 minutes |
| Human review effort per asset | Under 2 minutes average |
| Social media publishing prep automation | ~85% automated |
| Policy reference retrieval accuracy | Verified source reference every output |
- 90% Reduction in content creation time (weeks to minutes).
- Zero Hallucinations in production content due to strict RAG enforcement.
- Expanded Access for non-English speaking families through automated localization.
- Full Audit Trail ensuring every video frame is traceable to a specific policy clause.