Project Overview
Groopview, a real-time social-streaming start-up, wanted an AI avatar that could understand text and image in the middle of fast-moving group sessions. Avahi built an API for “Dualy,” for an event-driven AWS solution that orchestrates two Amazon Nova models—Nova Pro for precise reasoning and Nova Lite for rapid responses—to match each user query to the best service and reply in near-real time. The result is a scalable, low-latency experience that boosts audience interaction and unlocks new revenue streams.
About the Customer
As Groopview’s user base grew, hosts struggled to answer every question, surface contextual insights, and moderate chats quickly enough to keep viewers engaged. They needed an AI co-host that could:
The Problem
Liberty SF’s business development team previously reviewed thousands of court orders manually—a slow, errorprone process that limited the number of prospects they could contact and delayed time-to-offer. As filing volumes climbed, manual review threatened to bottleneck growth, increase labor costs, and allow competitors to reach claimants first.
- process mixed-media inputs (text, images) in real time,
- fetch answers from external APIs, and
- deliver sub-second interruption control for hosts.
Failing to solve these challenges risked higher churn and limited monetization.
Why AWS
Groopview selected AWS for its fully managed generative-AI stack and global, low-latency infrastructure. Amazon Bedrock provided turnkey access to the Nova family of LLMs—giving Groopview both speed (Nova Lite) and reasoning depth (Nova Pro) under a single API.
Why Groopview Chose Avahi
As a Premier-tier AWS partner with early-access experience on Nova models, Avahi combined reference architectures, accelerator libraries, and event-driven best practices to deliver a production-grade solution in just six weeks. Its ability to orchestrate multiple LLMs for cost-and-latency control set Avahi apart.
Solution
Avahi implemented a serverless, multi-layer architecture:
- Ingress & Control– Amazon API Gateway routes text and image streams to AWS Lambda for preprocessing.(
- AI Orchestration Within Amazon Bedrock, Avahi’s agent framework intelligently orchestrates calls to a collection of external APIs, ensuring low latency while preserving contextual continuity:
Nova Lite + Nova Pro for simple queries (<3 s average latency). Nova Lite + Nova Pro for complex queries (~7 s). Lambda merges model outputs, adds citations, and streams responses back to the app. - AI Orchestration– Within Amazon Bedrock, Avahi’s agent framework intelligently orchestrates calls to a collection of external APIs, ensuring low latency while preserving contextual continuity:
- Scalability & Observability– GPU-backed Amazon EC2 g6e instances auto-scale for bursty inference; Amazon CloudWatch provides unified metrics and alerts.
Key Deliverables
- Dual-model Nova orchestration (Nova Pro + Nova Lite)
- Real-time AI avatar with text and vision channels
- Context-management logic
- Event-driven ETL pipeline (Lambda → S3 → RDS)
- IaC, CI/CD runbooks, CloudWatch dashboards & alerts
Project Impact
Groopview launched Dualy on schedule and saw a surge in session stickiness and creator satisfaction. The dualNova design cut complex query latency by more than 80 percent versus Claude Sonnet, while preserving high accuracy and reducing compute cost per interaction.
- ~2.5 s average response for simple queries (Nova Lite and Nova Pro)
- ~7 s average response for complex queries (Nova Pro and Nova Pro) vs ~45 s on Claude Sonnet – 80% faster
- 2 fully supported interaction modes: text, and vision
- 6-week end-to-end delivery from kickoff to production
