Architecture Overview
YUIEN runs on a fully serverless Amazon Web Services (AWS) stack, combining Amazon Bedrock generative AI with Amazon S3 Vectors — AWS's native vector store — to deliver fast, source-grounded responses from a curated Jodo Shinshu corpus.
primary corpus
D.T. Suzuki, Unno, et al.
Shinran's writings
Generation Model: Amazon Nova Pro
YUIEN uses Amazon Nova Pro v1 via AWS Bedrock — AWS's flagship multilingual foundation model — for response generation:
- Fast responses: typically 2–4 seconds end-to-end
- Multilingual: native support for English, Portuguese, Japanese, Spanish, French, German
- Source-grounded: answers are constructed only from retrieved corpus passages
- Citation-aware: every response includes document attribution
Knowledge Base: S3 Vectors
YUIEN's corpus is indexed in an Amazon Bedrock Knowledge Base backed by Amazon S3 Vectors — AWS's purpose-built native vector store, launched in 2025. Compared to traditional vector databases, S3 Vectors provides:
- Pay-per-query economics: dramatically lower idle cost than always-on clusters
- Native AWS integration: no separate cluster, no provisioning, no scaling drama
- Durable storage: S3-grade 11-nines durability for embeddings
- Titan Embeddings v2: 1024-dimension semantic vectors with cosine similarity
yuien-docs
Primary curated PDFs and DOCX — IBS readers, BCA materials, scholarly translations.
shindharmanet
Web-crawled writings from shindharmanet.com — Suzuki, Unno, Bloom, Inagaki, et al.
shinranworks
Shinran Shonin's collected works — Kyogyoshinsho, Wasan, letters, in CWS translation.
384 indexed chunks across three curated data sources, all citation-attributed.
Serverless Infrastructure
AWS Lambda
Serverless API endpoint (yuien-api) — auto-scales, zero idle cost
API Gateway
RESTful HTTPS endpoint with CORS for browser integration
Amazon S3
Durable storage for source documents and vector embeddings
CloudWatch
Logging, performance metrics, and operational monitoring
Security & Privacy
- No conversation storage: questions are not persisted or logged
- IAM role-based access: least-privilege permissions on every AWS resource
- HTTPS encryption: all API traffic encrypted in transit
- No third-party tracking: no analytics, ads, or cookies
- Open source (planned): code will be publicly auditable
Cost Efficiency
The 2026 migration from OpenSearch Serverless to S3 Vectors reduced YUIEN's infrastructure cost dramatically — from a fixed monthly cluster fee to pay-per-query economics. This is what makes YUIEN sustainable as a non-profit Dharma service:
- Vector storage cost: reduced ~99% versus the prior architecture
- Pay-per-use: users only generate cost when actively querying
- Auto-scaling: serves 1 user or 1,000 with no code changes
- No upfront commitment: no reserved capacity, no minimum spend
Performance Metrics
| End-to-end response time | 2–4 seconds typical |
| Indexed knowledge base | 384 chunks / 384 documents |
| Supported languages | 6 (EN, PT, JP, ES, FR, DE) |
| API timeout | 30 seconds |
| Uptime target | 99.9% |
Roadmap
- Voice interface: ask questions via speech
- Mobile app: native iOS and Android
- API access: embed YUIEN in temple websites
- Filtered search: by author, text type, or time period
- Conversation memory: context across sessions
- Expanded corpus: more languages, more texts
Technical Questions?
Interested in the technical details, want to contribute code, or planning to deploy YUIEN for your temple? We're happy to share documentation and support your implementation.
Contact Technical Team