Applied intelligence lab · retail & commerce

We engineer the intelligence behind modern retail.

Zytun is an elite consulting and software lab. We design, build and run the AI, decisioning and distributed systems behind commerce at scale, for the largest brands and the fastest-growing startups.

An ancient olive tree with a twisting trunk and a broad canopy, the Zytun emblem
  • 01
    Retail-scale systemsHundreds of thousands of products, peak traffic, decisions at the edge in milliseconds.
  • 02
    Machine learning that shipsModels in production, every rollout measured against a control group.
  • 03
    Built inside Amazon and GoogleEngineers who built retail systems at the largest scale.

Selected work

A few examples from the workshop.

A sample from a much longer list of engagements. Client names are withheld; the problems were real.

Ask about more of our work
Global quick-service restaurant chain01

Real-time voice ordering at the drive-thru.

Menu orchestration across locations, voice filtering that isolates the guest from drive-thru noise, and an agent that turns speech into an order in real time, across hardware, software and ML. Plus purpose-built hiring software and pipelines to staff restaurants at scale.

  1. speech
  2. noise filter
  3. order agent
  4. menu
  5. POS
Industrial supply distributor02

B2B and B2C commerce, rebuilt end to end.

We re-platformed the entire e-commerce and tooling footprint, serving business buyers and consumers from one modern stack, along with the internal tools the teams run it with.

  1. catalog
  2. account pricing
  3. B2B + B2C
  4. ERP
Premium apparel retailer03

From breach to hardened, then modernized.

We recovered e-commerce systems from an organized international threat group, rebuilt them secure by design, then modernized retail operations with a new generation of point-of-sale.

  1. recover
  2. harden
  3. rebuild
  4. new POS
Stealth peer-to-peer marketplace04

A resale marketplace, from photo to sale.

We built the entire Rails application for a peer-to-peer startup preparing to launch: listing from a photo, buyer and seller flows, and the platform underneath.

  1. photo
  2. listing
  3. buyer
  4. ship
Retail decision platform05

Agents that merchandise e-commerce at scale.

We built the full decision engine, with multiple agents working together to merchandise e-commerce at scale and people approving what gets written.

  1. signals
  2. agents
  3. approve
  4. publish
Global performance athletic brand06

Segmentation that tunes itself for margin.

Marketing segmentation intelligence built from behavior and other commerce signals, with weights that auto-adjust through proprietary models to lift margin, conversion and CAC-to-LTV.

  1. behavior
  2. segments
  3. adaptive weights
  4. margin

Capabilities

Four disciplines, one senior team.

The problems between data, infrastructure and the customer, designed, built and shipped into production.

01 / systems

Distributed systems at retail scale

Catalog, pricing, inventory and order systems serving many storefronts and marketplaces at once. Event-driven, edge-deployed, built for peak traffic, and integrated with the PIM, ERP and feeds around them.

edge · events · PIM/ERP · marketplaces
02 / decisioning

Machine learning and real-time decisions

Propensity, lifetime-value and demand models. Price and offer decisions per customer, per session, in milliseconds, with a control group behind every rollout so results are measured, not attributed.

scoring · experiments · incrementality
03 / agents

AI agents for commerce operations

Agents for forecasting, buying, allocation, markdowns and campaign setup. Fine-tuned on your own data where that beats a general model, with a person approving before anything is written to the ERP or an ad platform.

agent design · fine-tuning · evals · approvals
04 / experiences

AI-native experiences at scale

Storefronts, apps, voice and internal tools rebuilt around intelligence: search that understands intent, conversational ordering, and operator consoles that put agents and data in one place.

voice · semantic search · consoles
How it runs

Every decision measured. Every agent accountable.

Decisions are served at the edge and logged against a control group. Agents draft; people approve; systems write. This is a sample trace from the kind of runtime we build.

zytun / decision-runtimesample
$runprice.decide --session 7f3a
→fetchprofile · cart · live inventory
→scorepropensity · LTV · demand
→decideoffer B (treatment)
→measurelift ledger vs. control
✓returnserved from the edge
 
$runagent.markdowns --dept apparel
→draftsell-through · open-to-buy
4 markdown proposals ready· awaiting buyerapprove
$ 

Clients

Enterprise depth. Startup speed.

The largest brands

Modernize what already runs the business.

Retrofit intelligence into catalog, pricing, inventory and order systems without stopping the store. Architecture reviews, build-versus-buy calls, platform cost reduction, and agents that work alongside merchants and planners.

The fastest-growing startups

Build the platform you will need at 100×.

AI-native products and the infrastructure under them, designed for the scale you are heading toward. A senior team that ships from week one, and fractional technical leadership while you hire.

AdvisoryArchitecture reviews, build-versus-buy, vendor selection and platform cost reduction. Fixed scope, written recommendations.
BuildA senior team delivers a working system, with measurement built in from the first release.
Embedded leadershipFractional or interim CTO: ownership of architecture, team, vendors and budget for a defined period.

Lab

Built here, running in production.

Products we build and run ourselves, most of them live at scale. What we learn running them comes back to our clients.

  • Real-time pricing and offer decisions

    Scores each customer and session, chooses the price or offer, and measures the result against a control group.

  • Sell anything from a photo

    Photograph an item; the service writes the listing, sells it, negotiates with the buyer and arranges shipping.

  • Merchandising agents

    Forecasts, open-to-buy, allocations and markdowns prepared for a person to approve, then written into the ERP.

  • Catalog content at scale

    Generates and checks product attributes, copy and images for large catalogs, with review before publishing.

  • Marketing spend measurement

    Measures what paid media actually adds using geographic and audience holdouts, and keeps a ledger of verified lift.

About

Zytun means olive.

The olive tree grows slowly, lives for centuries and keeps bearing fruit. That is how we think about the systems we build. Zytun is an applied intelligence lab of senior engineers and technology leaders who have spent their careers building commerce at scale.

We bring the same engineering standards to client work and to our own ventures, and we work as one senior team, on a few engagements at a time.

Amazon & Google
Engineers who built retail and marketplace systems at the largest scale.
CTOs & engineering leaders
People who have run technology at commerce software and e-commerce companies.
Senior, hands-on
The people you meet are the people who design and build your system.
  • 01Measure, don’t attribute.Every system we design keeps a control group and a ledger of verified results.
  • 02Primitives first.Small reusable pieces (identity, scoring, decisions, measurement) composed into products.
  • 03People approve, systems write.An agent’s work ends with a person’s approval and a write into the system of record.

Talk to us

Bring us your toughest problem.

The problems others call impossible are where we do our best work, and where our creativity shows. A senior engineer reads every brief and comes back with ideas, not a sales pitch.

  1. 01You send a short brief
  2. 02A working session on the problem
  3. 03A written proposal with our approach
You are
What do you need? Pick any
We only use this to reply to you. Privacy