NovaAssist CX : Multimodal Support Agent

Transform customer support with NovaFlow. An autonomous agent powered by Amazon Nova, handling complex logistics, CRM updates, and multimodal interactions with enterprise-grade reliability.

Comment

Inspiration

Customer support for enterprise logistics is a high-friction environment. Support agents are often forced to toggle between fragmented dashboards, CRM tools, and database queries while trying to diagnose hardware issues from customer-sent screenshots. We wanted to build a "Thinking Agent" that doesn't just chat, but acts. By leveraging the multimodal power of Amazon Nova, we built a system that can see what a customer sees and resolve issues autonomously.

What it does

NovaFlow AI is an autonomous customer experience platform that bridges the gap between multimodal intelligence and real-world execution.

Multimodal Vision: Nova analyzes customer-uploaded photos (like hardware damage or flickering screens) to authorize replacements or diagnose technical faults. Logistics Retrieval: It queries high-fidelity databases (DynamoDB) to verify order status and tracking details in real-time. Autonomous Fulfillment: The agent handles the "grunt work"—sending professional apology emails via AWS SES, updating HubSpot CRM records, and creating internal support tickets automatically.

How we built it

Intelligence Layer: Powered by Amazon Nova (via Bedrock) for reasoning, multimodal computer vision, and tool-use orchestration. The Nerve System: Built with FastAPI for a high-performance backend that manages agentic state and external API integrations. The Command Center: A premium, glassmorphic Next.js dashboard that provides a unified interface for both human supervisors and the AI agent. Cloud Foundation: Scalably deployed on AWS App Runner, utilizing S3 for asset management and DynamoDB for enterprise data persistence.

Challenges we ran into

Deploying a complex, consolidated Next.js and FastAPI stack onto AWS App Runner required deep-diving into multi-stage container builds. We had to optimize how Python dependencies are persisted across build layers to ensure the agent remained high-performance. Additionally, fine-tuning the agent's "loop" to handle complex, multi-step logistics tools required careful system prompting to maintain extreme reliability.

Accomplishments that we're proud of

We are incredibly proud of achieving a fully live, production-ready deployment on AWS. Seeing the agent successfully "see" a damaged product in an image and then immediately pivot to checking a real DynamoDB order table and updating a CRM record—all without human input—was a major milestone for us.

What we learned

Building with the Amazon Nova family taught us the power of multimodal models in solving real-world utility problems. We learned how to bridge the gap between abstract LLM reasoning and concrete database execution, and gained significant experience in orchestrating serverless container deployments for complex AI applications.

What's next for NovaAssist CX : Multimodal Support Agent

The next step is integrating real-time voice synthesis to allow customers to talk directly to Nova via phone. We also plan to build out predictive supply-chain analytics, allowing the agent to proactively reach out to customers before they even realize their shipment might be delayed.

Built With

Share this project:

Updates