Case study 01
Applied AI · Commerce
Kandid AI Commerce Platform
A Shopify app and merchant dashboard for running AI sales and support agents on storefronts, WhatsApp, and Instagram. Merchants can sync product and brand knowledge, configure agent behavior, review conversations, create broadcasts and automations, and track commerce activity.
My role
Architected and implemented the platform across the SaaS control plane, real-time communication services, embedded storefront widget, Shopify extensions, analytics, automations, and production infrastructure.
Core technologies
System path
A simplified view of the product workflow.
- 01
Shopper channels
- 02
Streaming agent runtime
- 03
Retrieval + memory
- 04
Commerce tools
- 05
Shopify + Meta
- 06
Analytics + operations
01
Product overview
What the platform does
The shopper experience runs across storefront web chat, WhatsApp, and Instagram. The channels share merchant knowledge and commerce capabilities while preserving channel-specific interactions, rich media, identity, and workflows.
The merchant side covers onboarding, authentication, roles, billing, catalog and knowledge synchronization, agent configuration, prompt and skill management, conversations, analytics, customer segments, broadcasts, and automations.
02
My work
Agent runtime and commerce features
- 01RAG and vector search over products, pages, files, FAQs, and brand knowledge.
- 02Persistent conversation memory and customer state across channels.
- 03Product recommendations, bundles, routines, cart operations, checkout, and order tracking.
- 04Text and voice conversations with customizable 3D avatars.
- 05Human-agent transfer with working hours and configurable notifications.
- 06Langfuse tracing, prompt management, and evaluation workflows.
03
Production
Deployment and operations
I operated and debugged the platform across its web applications, agent services, synchronization workers, browser services, databases, storage, and third-party providers. This included CI/CD, Docker, Linux, Nginx, cloud deployments, incident diagnosis, and production support.