automation

A ceramics shop didn't need Salesforce. It needed this.

Aug 7, 2026 · Valery — senior automation engineer

The before

The shop had a simple website. Visitors filled a form, the lead went to an email inbox, and that was the end of the story. Nobody filtered hot, warm, or cold leads. Nobody knew which inquiry was worth chasing in the next hour and which was just browsing. Leads got lost. Follow-ups didn't happen. The shop's team is not technical — they don't want to touch code, and they shouldn't have to.

The problem with off-the-shelf

I looked at the usual suspects: Salesforce, HubSpot, the local Israeli CRMs. Every single one failed on the same point — they're built for sales teams with dedicated admins, not for a small shop that lives on WhatsApp and wants its website to just work.

The shop needed three things no off-the-shelf tool gave them:

A website that actually captures and qualifies leads
A CRM their non-technical team can run
The ability to manage the entire store — products, texts, images, languages — without code

So I built both: a new storefront and a CRM, as one system.

Layer 1 — the storefront: an AI chatbot that qualifies leads

The new website has an AI chatbot. When a visitor shows interest, the bot asks a few smart questions, builds a lead from the answers, and hands it to the owner for follow-up. There's also a WhatsApp button for the business — because in Israel, that's how people actually buy.

Layer 2 — the CRM: AI sorts every lead

The CRM receives the leads from the site. Every lead — including the full conversation — goes to AI, which classifies it as hot, warm, or cold. Then the owner sees a drag-and-drop pipeline: new → in treatment → sale → failure. No training needed. If a lead moves to "in progress", the system automatically creates a customer record with everything captured — what they asked about, what they bought, what they didn't.

Layer 3 — managing the whole store without code

This is the part the shop owner actually loves. From inside the CRM they can:

Upload products manually
Import a whole catalog from PDFs — AI reads the PDF, splits it into individual products, the owner approves, and it goes live
Edit every text on every page of the website
Manage all images and page backgrounds
Add new product categories

And the site is trilingual — Hebrew, Russian, English. The owner writes in Hebrew; AI translates everything to the other languages automatically.

The stack

Because I believe in boring, working systems:

Storefront: Next.js 16, React 19, Tailwind 4, next-intl (he/ru/en)
Backend: FastAPI, SQLAlchemy, Postgres + pgvector, Alembic
Workers: Redis + arq
AI: OpenAI (chatbot + lead classification), Google Gemini (PDF extraction + image generation)
Storage: MinIO/S3
Auth: JWT, argon2, 2FA, RBAC (owner/admin/viewer)
Infra: Docker + nginx on a VPS

Where it stands

The system is built and live, and right now it's being tested by the shop in real use. I'm not going to give you fake numbers — the data is being collected as we speak, and I'll share the real ones when they're in.

What I can say: no lead has to get lost anymore, and the owner runs their entire online store without a developer.

The takeaway

Most small businesses don't need Salesforce. They need a system that fits how they actually work — WhatsApp-first, no-code, in their language. That's the kind of thing I build.

Have a process worth automating? Let's talk.

Want to see how this works in your business?

A short call, no obligation. We'll figure out where leads fall through — and make a decision.

Contact me