Platform

Eight surfaces,
one foundation model.

How each capability works, what it replaces, and what it is measured against. For the short version, see the homepage.

— — One substrate, eight surfaces

Index your catalog once.
Every capability rides the same vectors.

That is the economics of infrastructure: each surface below reuses embeddings you have already paid to compute. Add a capability, not a vendor.

Searchtext → product, image → productrank 1–2 on 9 of 10 benchmarks
Agentic Searchmulti-step, outfit and intent awareanswers, not result pages
Trend Intelligencehyperlocal — city and cohort levelmonths of early warning
TrendsFlow™trend lineage and lifecycleproprietary
Catalog Enrichmentattribute extraction at scale100+ attributes / SKU
Personalizationtaste vectors per shopper+20–40% AOV (modelled)
Size & Fitfit intelligence and returns riskattack the 30% return rate
Commercial Intelligenceassortment and demandgap analysis in plain English
Visual Similarityphoto → product, dedupe, street-to-shop#1 open model on LookBench
01 — One world model

One world model for fashion,
powering every product.

A multimodal foundation model trained on fashion's visual and commercial language — the layer the whole platform reasons on. Already state of the art on retrieval benchmarks against Marqo SigLIP and Google SigLIP2.

MODA — open on github.com/hopit-ai/MODA
What it understands
100+ Silhouette & construction Cut, fit, seams, garment grammar.
500+ Fabric, color & texture Material, drape, palette, print.
1M+ Trend context What's rising, peaking, fading & why.
20+ Demand & price signals Sell-through, elasticity, commercial fit.
What it powers
Section One

Agentic Search & Personalization

Section Two

Catalog Intelligence

03 — A catalog that enriches itself

A catalog that enriches itself.

Every SKU becomes an agent-friendly record — full attribute grammar, occasion tags, silhouette detail, visual keywords, and reasoning for why it will convert. Built for shoppers, search engines and downstream agents at once.

20M+

Trained on twenty million e-commerce product images to learn what actually converts — for which aesthetic, on which surface, to which buyer.

Trend attributes

What's rising, peaking, fading — stamped onto every SKU from Trend Lineage.

Visual attributes

Crop, composition, palette, styling — scored against what converts for this aesthetic.

Text attributes

Copy, tone, keyword — matched to the language that gets a buyer to checkout.

We write it The copy that converts.
We generate it The images that convert.
Before · table stakes Off-White Linen Shell Peplum Cami Blouse
Attributes per product — 20–30 granular Accuracy — 85% Taxonomy — flat list per category Custom taxonomy — generic apparel Languages — English only
After · best-in-class enrichment Off-White Linen Shell Peplum Cami Blouse

"Breezy linen cami with sculpted shell cups and a peplum flare — made for warm days, balmy nights, and effortless getaways."

occasion · brunch silhouette · fitted bodice fit · sweetheart fabric · linen wear · warm-weather visual · white cami visual · peplum drape buyer · resort
Attributes per product — 100+ granular Accuracy — 90%+, professionally auditable Taxonomy — 6-level hierarchy · brand-specific Business outcomes — conversion, revenue, keyword coverage Languages — multi-language SEO & local phrasing
+20–40% AOV lift from richer PDPs
+60–85% Zero-result recovery on search
+15–30% Organic revenue from SEO / AEO
2–5× Content throughput per merchandiser
Section Three

Fashion Ontology

04 — A schema LLMs can reason on

Fashion data,
legible to LLMs.

Under every product and trend is a canonical schema — silhouette, occasion, fabric, palette, cohort, price, lineage. It is what turns a messy catalog into structured intelligence that shopping agents, AI search engines and downstream models can act on. 

// A single SKU as agent-consumable structured intelligence
{
  "sku_id": "LNN-CAMI-01",
  "name": "Off-White Linen Shell Peplum Cami Blouse",
  "silhouette": {
    "family": "top/blouse",
    "shape": "fitted-bodice",
    "neckline": "sweetheart",
    "strap":    "spaghetti"
  },
  "fabric":   { "material": "linen", "drape": "crisp" },
  "palette":  { "family": "neutral", "tone": "off-white" },
  "occasion": ["resort", "brunch", "day-out"],
  "trend_lineage": {
    "parent_macro": "Neo-Victorian Romanticism",
    "cluster":      "Organza Floral Applique",
    "call":         "strong-buy",
    "score":        62
  },
  "cohorts": ["clean-girl", "quiet-luxury"],
  "price_band": "mid"
}

What every downstream agent
needs — and rarely gets.

Canonical taxonomy across silhouette, fabric, occasion, palette and cohort.
Trend lineage stamped onto every SKU — the whole graph, not a label.
Consumable by LLMs, MCP tools and agentic shopping surfaces out of the box.
Reusable across your PIM, PLM, DAM, search stack and marketing agents.
Section Four

Trend & Lineage Intelligence

05 — Every trend, scored and ranked

Every trend, scored and ranked
before you commit.

Not a mood board. Every micro-trend surfaces as a scored cluster of visual atoms with a saleability call, a lifecycle read, and the runway and social signal behind it.

Macro Gorpcore
M04_010 · 119,508 atoms
Strong buy · 62 Trendy
1–2mo
62
6mo
58
1yr
50

Still climbing — strong continuing trend.

Macro Bohemian Revival
M04_013 · 11,202 atoms
Hold · 56 Stable
1–2mo
56
6mo
47
1yr
39

Established and accelerating.

Macro Blokecore
M04_003 · 6,239 atoms
Strong buy · 65 Trendy
1–2mo
65
6mo
62
1yr
54

At peak — expect cool-off.

Micro Washed Denim Utility
U04_1404 · 1,425 atoms
Hold · 54 Dormant
1–2mo
54
6mo
47
1yr
40

Foundational mass-market staple.

Trends don't appear. They evolve, split and merge — and the lineage is the read.

Parent macro
Neo-Victorian Romanticism
Merge · 2024
Current cluster
Organza Floral Applique
2,437 atoms
Split · 2026
Beaded Organza Lace Mesh
Split · 2026
Spectral Romanticism
Evolve · Continuity Split · Diverges into Merge · Converges with

The same lineage layer feeds every decision downstream.

Draws from · Saleability Commercial calls

Buy / hold / avoid and assortment-gap calls.

Draws from · Live trend graph Personalization

Ranks each shopper against the live trend graph.

Draws from · Rising clusters Growth marketing

Briefs campaigns off clusters before they peak.

Draws from · Atom embeddings Agentic search

Retrieves on aesthetic intent, not just keywords.

Section Five

Commercial Intelligence

06 — Ask the engine anything

Ask the engine where your assortment is thin
and where the opportunity lies.

Commercial intelligence in plain language. The engine reads your catalog, your competitors', and the live trend graph — and hands back a brief the merchandiser can sign off in a meeting.

In linen dresses, what micro-trend opportunities are we missing, based on competitor assortment gaps?

Thought for 7s

The biggest whitespace in linen dresses is around versatile summer day silhouettes — midi lengths, sleeveless shapes, and waist-defined constructions show the largest competitor lead. The strongest SKU gaps are functional day-dress shapes, then silhouette and neckline refinements where competitors still offer deeper choice. 8 opportunity themes identified; largest gap is 62 SKUs in linen midi day dresses.

Trend group #1 · surfaced

Strap-Detail Linen Dresses

Competitors: 52 · You: 26 — 26 product gap
SilhouetteStraight · Drop · A-line
PaletteMuted linen · fondant gold · classic midnight
SignalsSpaghetti · wide straps
BuyerDay dress · destination · everyday
Section Six

Hyperlocal Trends

07 — Not one trend graph, many

Mumbai's blokecore
is not Milan's.

Global trend graphs are a starting point, not an answer. Hopit resolves the world model at city and cohort level — so a merchandiser in Delhi and a merchandiser in New York get different calls off the same schema.

Delhi street-style — an autumn drape reading against a City Journal masthead

Delhi · autumn drapes reading against a marigold palette — a signal the global model doesn't see on its own.

Mumbai · IN

Blokecore ascendant among Gen-Z men — sports-glam palette breaking into fast-fashion.

Instagram 30d+62%
Search intent+48%
Global baseline+22%
Delhi · IN

Fusion Indo-western dressing merging with quiet-luxury minimalism — palette shifts toward stone and champagne.

Weekend event demand+41%
Repeat-buyer share34%
Global baseline+11%
Bengaluru · IN

Techwear utility on the rise — sneaker-adjacent, muted tones, high add-to-bag from the Gen-Z DTC cohort.

Add-to-bag lift+27%
Session length+18%
Global baseline+7%

Every product, every cluster, every cohort resolves against a city-specific view of the world model. Wedding season in Chennai, festival demand in Dhaka, prom pull in Chicago — the merchandiser sees the read they can actually act on.

08 — Custom fashion tech & AI

When you need something bespoke,
we build it on the foundation.

The world model gives us a running start. Every custom system we ship — planning co-pilots, forecast APIs, size-and-fit models, returns-risk scoring — inherits it. You get months of head-start over a from-scratch build.

Demand-forecast API Saleability scores piped into your planning stack.
Auto-merchandised PLPs Category pages reordered by live trend & margin.
Designer co-pilot Lineage-aware ideation inside your design tools.
Season assortment planning Assortment planning co-pilot for the upcoming season.
Cross-sell & upsell APIs Higher AOV using cross-sell and upsell.
Size & fit Proprietary scoring for size and fit recommendations.
Returns-risk scoring Fit-risk flags at the point of add-to-bag.
Your idea here A four-week pilot, then a proper build. Anchored on the foundation.

MODA · World models · Embeddings · NER · Causal trend lineage · Generation

09 — Already running with paying customers

It's already running
in production.

Four product lines are live with paying customers today — not a roadmap. Advanced scoping conversations underway with a handful of the industry's largest marketplaces and content platforms.

Live with paying customers
Shipped Chumbak

Catalog intelligence and merchandiser-facing trend and commercial intelligence across the assortment.

Shipped The Label Life

Trend lineage, commercial intelligence and enriched catalog feeding the merch team.

Shipped Outzidr

Catalog intelligence and discovery across a fast-moving apparel assortment.

Shipped Ozi

Catalog intelligence and search enrichment tuned for a D2C apparel marketplace.

In deployment
A Fortune 500 marketplace Agentic search across a catalog in the hundreds of millions of SKUs. In deployment
A top-10 US fashion retailer Trend intelligence and assortment planning across the private-label book. In deployment
— — The lab underneath

Don't take our word.
Take our benchmarks.

Our models are measured in public — full corpus, one harness, competitors included, losses shown. We publish the experiments that failed, too. Consider it vendor diligence we already did for you.

SurfaceResultMeasured against
Image search (LookBench)#1 open · 67.63above a 1.24B model
Text search (10 cells)rank 1–2 on 9SigLIP-SO400M 878M, ZooClaw
Under 250M classleads 5 of 6catalog and caption suites
Full benchmark tables → Open models on Hugging Face → Research notes →
Ship log
Aug 10MODA Pro Lite released — 213M open-weights fashion encoder
Aug 09Full-corpus benchmark across six retrieval systems published
Aug 03Trend substrate v07 — 21 macro trends re-validated
Jul 31Occasion and mood axes across 3.19M catalog items
Jul 26Retrieval-head experiment — published as a negative result
10 — Two ways to start

Off the shelf,
or built on the foundation.

Take what's already running, or have us build the specific system your problem needs. Either way, you're on the same infrastructure — the difference is how much gets built for you.

Path One

Off the shelf.

Commercial, Catalog or Agentic Search — configured onto your stack, not built from scratch.

1–2 weeks Onboarding — data connected, taxonomy mapped, live.
Path Two

Built on the foundation.

A bespoke system for a problem the shelf doesn't solve — built by our Forward-Deployed Engineers, in your codebase.

2 weeks Understanding — we scope the problem, in your data.
6–16 weeks FDEs onsite — solve and deploy, depending on the problem.
Ongoing We move out. Support continues.
Book a call Fixed fee · two weeks · your catalog, your metrics — you keep the evaluation report either way.

The data covenant

  • Your catalog never trains our public models.
  • VPC and dedicated deployment available.
  • Deletion on exit, in writing.

The question every retailer should ask an AI vendor — answered before you ask it.

11 — Talk to us

One meeting. See if it fits.

We respond personally to every inquiry. Book a call and we will bring specific examples from your category, along with a proposed pilot shape.

Book a call MODA on GitHub Read the research