Customer Story
mycrux. uses the Cruxy API across stylist recommendations, fraud detection, and seller operations - from day one, in eleven Indian languages, on UPI-native billing.
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.
The challenge
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.
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.
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
mycrux. routes each workload to the model built for it - not the most capable model for everything.
Cruxy Stylist
Vaanimycrux.'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
MiraCruxy 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
KaviInside 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.
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
Hindi, Telugu, Tamil, Bengali, Marathi, Kannada, Malayalam, Gujarati, Punjabi, Urdu, English. No add-on, no translation layer.
Cruxy bills mycrux. the same way mycrux. bills its users. No INR ↔ USD conversion friction, no surprise FX losses on the API line item.
Predictable monthly costs for an India-first business. mycrux.'s CFO doesn't have to model rupee weakness into the AI line.
GST, festival calendar, monsoon, IRCTC, Indian law basics. Not prompt engineering - model-level.
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