Enabling Real-Time Conversational Commerce for Interloop LLC on AWS

Customer Overview

Interloop LLC is an emerging AI startup focused on revolutionizing how people buy online through Agentic Commerce. The company is developing a new category of experience where users interact with animated AI avatars to explore products and complete purchases by matching shopping preferences with product sentiment through conversation.

Interloop LLC was spun off from AMGI Studios as an innovation paradigm shift in online marketing: why spend for the impression when you can pay for the match?

Built on Amazon Web Services, the platform combines real-time voice interaction, computer vision, and behavioral analytics to create a personalized commerce journey that evolves with each session. Conversational interactions, behavioral signals, and computer vision data are aggregated to support a Customer 360 approach focused on increasingly personalized recommendations over time.

Key Challenges

Interloop LLC sought to build a seamless speech-to-speech experience where users interact with an avatar to explore products and complete purchases through conversation. 

Several challenges emerged:

• Avatar response latency initially ranged from 1 to 8 seconds, disrupting conversational flow
• Multiple systems had to be orchestrated across voice, LLM inference, catalog lookup, and personalization
• The platform needed to support a US-first rollout with a path to global expansion
• Evaluating voice-to-voice and text-to-speech models for voice quality, latency and flexibility to tune lip-sync performance, and cost optimization.

The objective was to reach a response time between one and three seconds while preserving voice quality and enabling a flexible architecture (one second was achieved).

Solution

Solution Architecture

TrackIt partnered with Interloop LLC to evaluate their existing proof-of-concept AI stack, which relied on third-party services such as ElevenLabs, OpenAI, and Groq, and evolve it to a modern, fully AWS-native architecture. 

Latency Profiling and Provider Evaluation

A detailed latency profiling exercise identified bottlenecks across model inference, audio generation, API layers, and product lookup workflows. Based on these findings, multiple providers were evaluated and a data-driven recommendation strategy was established.

The evaluation compared providers including Deepgram, ElevenLabs, and Nova Sonic across speech latency, session bootstrap time, and overall user-perceived responsiveness. Benchmarking revealed significant differences in response timing and initialization overhead, helping guide architectural decisions around conversational performance, orchestration flexibility, and scalability. 

While Deepgram and ElevenLabs demonstrated faster first-audio response times in several scenarios, the evaluation highlighted broader trade-offs across orchestration flexibility, operational overhead, scalability, and cost profile. These findings helped guide the decision to standardize the conversational orchestration layer on Amazon Nova Sonic and Amazon Bedrock AgentCore Runtime as part of a more integrated AWS-native architecture. 

Building the Speech-to-Speech Architecture

The conversational flow was restructured using Amazon Nova Sonic as the orchestration layer. This enabled real-time voice interaction management and the invocation of backend services through structured tool calls.

The speech-to-speech layer was deployed using Amazon Bedrock AgentCore Runtime. The model orchestrates interactions by triggering backend services for product discovery and transaction workflows, integrating with systems such as digital fingerprint services, MCP catalog, and computer vision data. 

Session and interaction data were streamed through Amazon Kinesis Data Streams and processed using Amazon Flink, enabling real-time enrichment of user signals. These signals were further modeled using Amazon SageMaker to refine recommendations and user profiling. 

Interloop LLC’s development tooling was migrated to Amazon Bedrock, providing centralized model access, governance, and cost control, along with the flexibility to switch between providers. Guardrails were implemented to ensure all interactions remained focused on commerce use cases.

Innovation Highlight

Interloop LLC’s platform introduces an agentic commerce model where interaction, personalization, and transaction are unified into a single conversational flow.

Users engage with an avatar through natural speech. Behavioral signals, preferences, and facial expression data are captured in real time. These signals are processed through an analytics pipeline that continuously refines recommendations based on evolving user context, supporting a Customer 360 model centered on increasingly personalized interactions over time.

This creates a system that adapts dynamically during each session, rather than relying on static recommendation logic.

Impact

The engagement established a foundation for a more responsive and scalable platform:

• Projected reduction in response time from up to 8 seconds to 1 to 3 seconds
• Improved conversational continuity and reduced interaction friction
• Centralized AI architecture with governance and cost visibility
• Flexibility to evaluate and switch between AI and voice providers
• Real-time data pipeline enabling continuous personalization improvements

A New Approach to Conversational Commerce

Interloop LLC’s platform moves beyond traditional e-commerce patterns by treating interaction as the primary interface. Rather than separating discovery, evaluation, and purchase into distinct steps, these actions are unified within a continuous conversational flow.

This approach enables a more natural form of engagement, where user intent, behavior, and real-time signals are interpreted together to guide decisions. Over time, this creates a system that becomes more context-aware with each interaction.

By combining agent-based orchestration with real-time data processing, Interloop LLC introduces a model that shifts commerce from navigation-driven to interaction-driven experiences.