Jivesh Lakhani
← Back to selected work

Case study 01

Applied AI · Commerce

Kandid AI Commerce Platform

Kandid is a Shopify app that puts an AI sales assistant on a brand's website, WhatsApp and Instagram. Merchants sync their catalog, tune how the agent talks, read conversations and set up broadcasts and automations. By July 2026 it was live on 40+ stores and seeing roughly 2M storefront sessions a month.

My role

I built most of it: the merchant dashboard, the agent backend and real-time messaging, the storefront widget, the Shopify extensions and the infrastructure.

Core technologies

Next.jsReactTypeScriptNode.jsPostgreSQLOpenAIAzure OpenAIAzure SpeechLangfusePineconeDrizzle ORMSocket.IOShopifyMeta Graph APIWhatsApp Cloud APIInstagram Messaging APIDocker

System path

A simplified view of the product workflow.

  1. 01

    Shopper channels

  2. 02

    Streaming agent runtime

  3. 03

    Retrieval + memory

  4. 04

    Commerce tools

  5. 05

    Shopify + Meta

  6. 06

    Analytics + operations

01

Product overview

What it does

Shoppers can talk to the same agent on the website, WhatsApp or Instagram. It knows the same products and can do the same things on every channel, but each channel still behaves like itself: WhatsApp gets interactive replies, the website gets product cards and voice.

Merchants get onboarding and billing, catalog and knowledge sync, agent settings, conversation history, analytics, customer segments, broadcasts and automations.

02

My work

The agent and what it can do

  • 01RAG over the store's products, pages, files, FAQs and brand docs.
  • 02Memory that follows a customer across channels.
  • 03Recommendations, bundles and routines, plus cart actions, checkout and order tracking.
  • 04Text and voice chat with a customizable 3D avatar.
  • 05Handoff to a human that respects the store's working hours and notifies the team.
  • 06Instagram post analysis (images and video) and AI comment moderation. Merchants choose per category whether to ignore, hide, reply publicly or follow up in DMs.
  • 07Journeys with triggers, branches, delays, webhooks and WhatsApp messages.
  • 08Langfuse for tracing, prompt management and evals.

03

Scale

Numbers

  • 0140+ Shopify stores with the app installed (July 14, 2026).
  • 02Over 40.8 million storefront sessions tracked since April 2024, about 2 million a month recently.
  • 0320,000+ shopper conversations with the agent across web, Instagram and WhatsApp in the 30 days before July 14, 2026.

04

Kandid CMS

A CMS on Shopify Metaobjects

A smaller product next to the main app: a CMS that lets merchants manage the data our agents use, stored as Shopify Metaobjects. I did the architecture, the Shopify integration, the admin UI and the APIs.

05

Production

Running it

I deployed and looked after 10+ production services for Kandid: the web apps, agent services, sync workers, browser services, databases and storage. Day to day that meant CI/CD, Docker, Nginx, Linux servers on three clouds, and debugging live issues while merchants were waiting on a fix.