Cruxy
Customer Story

Customer Story

mycrux. runs an India-first fashion marketplace on Cruxy.

mycrux. uses the Cruxy API across stylist recommendations, fraud detection, and seller operations - from day one, in eleven Indian languages, on UPI-native billing.

mycrux

3

Cruxy models in production

11

Indian languages live

<400ms

Mira fraud-check latency

About mycrux.

mycrux. is an India-first fashion marketplace covering both the consumer side (mycrux.in) and the seller side (business.mycrux.in). It pairs a 3D body-mapping virtual try-on with an AI personal stylist, fraud detection on every review and listing, and an AI copilot for sellers who don't have ops teams. mycrux. is built in India, for India - vernacular by default, GST-aware, festival-aware, monsoon-aware.

Fashion
B2C marketplace
B2B seller portal
Built in India
Genesis stage

The challenge

What mycrux. needed from an LLM provider.

Vernacular at scale

Indian shoppers don't all speak English. Sellers don't either. Generic LLM APIs treat Hindi, Telugu, Tamil, and Bengali as second-class - higher token costs, weaker reasoning, no understanding of register or honorifics. mycrux. needed first-class Indian-language support, not a translation layer.

Trust gap in fashion e-commerce

Fake reviews, fabric mismatched against descriptions, sizing fraud, color-shifted product photos. A marketplace at scale needs fraud detection on every review and every listing - which means the per-call cost has to be in paise, not rupees, and latency has to be sub-second.

Seller operations are messy

Small Indian brands juggle inventory, GST filings, courier partner selection, ad spend, and demand forecasting without dedicated ops teams. mycrux.'s seller copilot needs an LLM that understands Indian commerce - not one that thinks 'tax season' is in April.

The solution

One API, the right model for each job.

mycrux. routes each workload to the model built for it - not the most capable model for everything.

Cruxy Stylist

Vaani

Personal stylist

mycrux.'s personal stylist reasons over weather, occasion, skin tone, and India-specific context - festival, season, region - to recommend an outfit. Vaani's 200K context window holds a user's full preference history and recent orders. Vernacular replies in the user's language of choice. Balanced cost and capability for the everyday recommendation path.

Cruxy Score

Mira

Fraud detection

Cruxy Score runs fraud detection on every review and every product listing - fake reviews, mismatched fabric specs, color-shifted photos. Mira's sub-400ms latency means scoring happens in real time as listings go up. At ₹13 per million input tokens, the math works for high-volume per-call inference.

Cruxy Mode

Kavi

Seller copilot

Inside business.mycrux.in, sellers run their stores in Cruxy Mode - Cruxy proposes inventory moves, GST document generation, courier partner choices, demand forecasts, and ad-spend reallocations. Sellers approve each decision; everything is audit-logged. Kavi's 250K context holds a full seller dashboard, and its reasoning depth handles the multi-step planning these tasks need.

Behind the scenes: deterministic-first, LLM where it matters.

mycrux. ingests catalogs from external suppliers - raw Shopify-formatted tag exports, HTML body content, mismatched schemas. The naive approach is to throw every product at an LLM and pay per token. mycrux. doesn't.

A deterministic extractor parses structured tag patterns first - Fabric_* keys, regex over body_html, the obvious wins. Only the gaps route to Mira for LLM enrichment. On a recent ingest of 2,355 Suta products, 2,262 needed zero LLM cost. The remaining 93 went to Mira at ₹13 per million input tokens - small enough to be a rounding error.

"The right model for the right job" isn't a marketing line. It's how a margin-thin marketplace stays alive.

Built for India

Why Cruxy fit mycrux.'s stack from day one.

26+ languages, eleven live in mycrux.

Hindi, Telugu, Tamil, Bengali, Marathi, Kannada, Malayalam, Gujarati, Punjabi, Urdu, English. No add-on, no translation layer.

UPI-native billing

Cruxy bills mycrux. the same way mycrux. bills its users. No INR ↔ USD conversion friction, no surprise FX losses on the API line item.

INR-primary pricing

Predictable monthly costs for an India-first business. mycrux.'s CFO doesn't have to model rupee weakness into the AI line.

India context, built in

GST, festival calendar, monsoon, IRCTC, Indian law basics. Not prompt engineering - model-level.

Jan 2026 knowledge cutoff

Current, not stale. Models know about the 2025 GST council changes and the latest courier partners.

"We tried wrapping foreign LLM APIs and the math never worked. Vernacular cost three times more in tokens, latency was abroad-routed, and 'understanding India' meant pages of system prompt. With Cruxy, we just call the API."

- Srisanth, Founder, mycrux.

mycrux. on Cruxy

Stack at a glance.

Models in productionKavi · Vaani · Mira
Primary use casesStylist, fraud detection, seller copilot
Indian languages live11
Avg fraud-check latency (Mira)<400ms
BillingUPI, INR-primary
IntegrationDirect API
Visit mycrux.in →