Zach Naim · Product Leadership & Business Systems

— 01

Custom Sales OS ERP

Live Quote to cash

Eleven European manufacturers, 10,000+ products, and a sales process that ran on scattered price lists from first quote to final payment.

Year2026

RolePM · Architect · Builder

StackCloudflare Workers + R2, Airtable → Supabase, Shopify Liquid, Resend, pdf-lib, Cron

// custom-sales-os.spec

The catalog had to be a selling tool, not a reference10,000+ products across eleven European manufacturers, each pricing in its own format. It became a configurator that produces branded, bilingual quotes with packshots, rather than a lookup table. Started in Airtable so the studio could correct its own data from day one, then moved to Supabase Postgres when the catalog outgrew the row limits.
One click from quote to order, deliberatelyAn accepted quote converts into a customer order without re-entry, because every manual re-keying between the two was a chance to sell a price nobody would honour. Orders then move Draft → Submit → Update behind a Cloudflare Worker that keeps supplier credentials server-side, and a finalize flag means only a deliberate submit reaches a manufacturer.
The sale does not end at the orderPayment tracking, invoicing and product status live in the same system, with confirmations generated as PDFs and sent by email or WhatsApp. A daily job flags orders 50 days from arrival and emails the owner, because the recurring failure was never missing data, it was nobody looking in time.

10,000+ products from eleven manufacturers in one priced catalog, and a straight line from quote to order to payment. The studio stopped assembling sales out of scattered price lists and started running them from one screen.

One person, start to finish. The same constraint that made it fast would have broken it at twice the scope.

— 02

Siri Bridge

Shipped iOS App Personal build

The smart-home devices Israeli homes actually run ignore HomeKit, because HomeKit ignores them.

Year2025 – 2026

RolePM · Architect · Builder

StackSwift, SwiftData, AppIntents, NSUserActivity, Keychain, XCTest

// siri-bridge.proto

Reverse-engineering was the only path, and a fragile oneSwitcher publishes no API, so the protocol was recovered from packet captures of its own app. The alternative was waiting for an official integration that may never ship. A firmware change can break this at any time — a risk worth taking on a personal build, and not on a client's.
Local first, cloud only as fallbackThe vendor's own cloud relay rejects roughly half of all requests. Control runs over local TCP with CRC-CCITT signing when the phone is on home Wi-Fi, and falls back to the encrypted cloud path with retries sized for a server outside my control.
Tested like the failure is expensiveCredentials sit in Keychain and 65 XCTest cases run before anything ships, which is heavy for a side project until you remember the device is a water heater. Bilingual from the first commit, because retrofitting Hebrew is never cheaper later.

Voice control for devices that never supported it. A personal build, kept deliberately small: the point was finding my own limit with no team and no spec to hide behind.

The engineering was the easy part. Deciding how much fragility to accept, and where, was the actual work.

— 03

Website Review & Refactor

Ongoing Product Ops

A Shopify storefront taking real orders, maintained like a marketing site.

Year2026 · 2 weeks

RolePM · Architect · Builder

StackShopify Liquid, Python, Admin API, PageSpeed, Translate & Adapt

// website-review.scope

Process sized to the downside, not the stackEvery release ships with a one-click rollback backup, a separate staging theme, and a changelog of each file touched. Unusual for Shopify Liquid work, and justified not by engineering taste but by the loss: a bad release here costs orders, not tickets.
A months-long refactor shipped as one reversible unitThe CSS rework spanned weeks, so it was bundled into a single named version rather than released piecemeal. Dripping it out would have made rollback impossible without discarding months of cleanup. Reversibility mattered more than release cadence.
Every caught bug becomes a permanent gateSite-health, performance and template scans run before each release. When a check catches something real it becomes a required step, so the audit suite grows from actual failures rather than imagined ones.

The storefront went from something changed cautiously and rarely to something that ships weekly with a known way back.

Not fancy. The discipline of version control, rollback paths, previews and written decisions, applied to a stack that didn't ask for it.

— 04

Email Builder

Live · v2 Internal tool

Every newsletter had to pass through someone technical. That person was me.

Year2026

RolePM · Architect · Builder

StackShopify Liquid, vanilla JS, Cloudflare Worker, Shopify Files API

// email-builder.scope

Fewer choices, on purposeA five-section form generates ActiveTrail-ready HTML in one click. A free-form editor would have been easier to build and would have let the brand drift within a month. Constraining the tool was the product decision; the rest is implementation.
AI drafts, a person decidesAn LLM drafts copy from a reference URL, the product images and an editable tone prompt. It produces a starting point rather than a final send, because the failure mode of automated copy is not bad grammar, it is confident brand-wrong writing nobody catches.
Assets read from where they already liveThe picker reads the Shopify product and file libraries directly instead of accepting uploads. Every manual re-upload is an opportunity to publish last season's price to the entire list.

Newsletter production cut by more than half. A content lead with no technical background now drafts, previews and ships without opening the platform editor, and without waiting for me.

Built so the brand voice never drifts, and so shipping a newsletter never waits on someone technical.

— 05

Translation Toolchain

Live Pipeline

793 products needed Hebrew. Machine translation would have been fast, and wrong about every brand, designer and collection name.

Year2026 · 2-day v1

RolePM · Architect · Builder

StackPython, Shopify GraphQL, OAuth, YAML, JSONL

// translation-toolchain.spec

Reversible by default, because the asymmetry is brutalEvery phase runs as a preview first, and each write is preceded by a JSON snapshot of the prior state. An unwanted change means correcting 793 products by hand; confirming costs four seconds. When costs are that lopsided, the default should never be the fast one.
The owner's decisions, written downA 370-term glossary fixes which brand, designer and collection names stay Latin in Hebrew, and 18 collections are marked untouchable. These are commercial decisions, not linguistic ones, so they belong to the owner and are recorded rather than inferred by a model.
Writes that are safe to repeatTranslations are applied through Shopify's GraphQL API and are idempotent, so a failed batch can simply be re-run instead of reconciled by hand. 1,750+ writes shipped in iteration one on that basis.

793 products localized in two days, against a manual alternative measured in dozens of hours. The 370-term glossary means the same call never gets argued twice.

When a missing closing tag on three product pages broke the Hebrew PDP layout, the fix took ten minutes. The HTML-tag-balance audit step that would have caught it preemptively took an hour, and now runs on every cycle. That's the whole job.

I product.

About

Selected Work

Custom Sales OS ERP

Siri Bridge

Website Review & Refactor

Email Builder

Translation Toolchain

Experience

Independent, Product + AI Systems

Client Advisor & Account Manager

Product Operations Manager

Product Manager

Mobile Product Manager

Product Manager, Mirrorball Slots

Product Manager

Contact