SAINI, AI Infrastructure for Emotional Intelligence

Building an AWS Bedrock-powered emotional intelligence engine that delivers daily trauma-informed AI reflections, proactive nudges, and real-time human support across health, justice, and workforce.

Comment

Inspiration

We were inspired by the silent crisis of trauma-driven instability affecting millions navigating justice re-entry, mental health recovery, and workforce reintegration. While billions are spent on disconnected systems, people still fall through the cracks. We asked: What if emotional intelligence could become infrastructure? That question gave birth to SAINTE Agent Core — an AI-powered, trauma-informed conversational engine that builds daily human momentum through reflection, empathy, and measurable progress.

What it does

SAINTE Agent Core enables users to check in emotionally each day through text or voice. Using Amazon Bedrock, it interprets tone, classifies emotional tier, and responds with an empathetic nudge or reflection. Each interaction is logged, vectorized with Titan v2 embeddings, and stored as long-term “memory,” allowing the agent to learn a user’s emotional patterns over time. A scheduled EventBridge workflow automatically re-engages inactive users, ensuring continuous presence and care.

How we built it

We designed a multi-region serverless architecture using: us-east-2 (Ohio) for core Lambdas, DynamoDB, and Titan/Nova models us-east-1 (Virginia) for Claude 3 Sonnet responses All compute runs through AWS Lambda micro-services: check_in_handler → orchestrates incoming user reflections respond_nudge_us_east_1 → generates trauma-informed responses with Claude 3 Sonnet update_memory / retrieve_memory → build and query Titan v2 semantic vectors auto_nudge_runner → daily scheduled nudges for inactive users analytics → real-time behavioral and sentiment analytics Data is securely stored in DynamoDB, visualized in a Streamlit dashboard via API Gateway.

Challenges we ran into

Balancing empathy with brevity in generated responses. Synchronizing embeddings and tier classifications across distributed regions. Handling DynamoDB float precision for embedding storage (solved with Decimal conversions). Ensuring safe, consent-based escalation pathways for higher-risk users.

Accomplishments that we're proud of

What we learned

Building emotionally aware AI requires more than sentiment — it’s about continuity, pattern recognition, and tone modulation. Cross-region Bedrock orchestration adds resiliency and performance benefits for multi-LLM architectures. AWS’s modular services make it possible to deploy a real-time, ethical agent system with minimal ops overhead.

What's next for SAINI, AI Infrastructure for Emotional Intelligence

We’re extending Saini’s capabilities with vector memory search, personalized resilience scoring, and federated learning for privacy-preserving model adaptation. Our next milestone is deploying a full Bedrock-powered reflection journal with peer support dashboards integrated into workforce and Medicaid systems.

Built With

Share this project:

Updates