First-principles retail
Features come from paid reality and browsing truth—not vanity charts—so scores stay tied to money-moving events.
Learning layer
Understand customers at the speed of commerce
Each checkout sharpens your picture—RFM paths, propensity scores, country-level movers and optional Grok-authored narratives. Anticipate peaks, catch softness early and merchandise with evidence—not guesswork.
[ At our core ]
Signal from every purchase
Postgres stays the system of record while a MongoDB feature store learns behaviour across three tiers: shoppers, each store, and whole markets. Models run where your stack already lives.
0
Behaviour tiers
Customer · Store · Country
0+
Model outputs
CLV, churn, forecasts & more
0h
Batch-friendly AI
Grok analyst cadence
∞
Embedding depth
OpenAI vectors · RAG-ready
[ How we think ]
Scale without rewiring
One architecture serves a boutique and a multi-country marketplace roll-up; scope grows with your catalog.
Ship insights nightly
Scheduled jobs plus live hooks keep layers fresh; logs and hooks capture what the AI stack did overnight.
[ Infrastructure ]
Industrial AI, merchant-calm UI
Vercel AI SDK on the surface; xAI Grok batches, Anthropic Claude, and OpenAI embeddings—all tenant-scoped.
Vercel AI SDKxAI GrokAnthropic ClaudeOpenAIMongoDBProphet
[ Capabilities ]
What lands in your console
No separate “data lab”. Everything routes next to orders, catalog and payouts.
Forecasts
Demand & revenue outlook
Store outlooks, next-purchase timing and lifetime-style signals to pace promos, staff and stock.
Risk
Churn & spike radar
Probability scores plus statistical callouts surface softness or demand spikes before they surprise you.
Merch science
Basket & category intelligence
Association rules and category movers show bundles and trends—not isolated SKUs in a vacuum.
Marketplace
Cross-vendor learning
Buyer roll-ups, vendor behaviour snapshots and cached recs when you sell inside Etrovvo’s shared cart.
Moments
Short-term behaviour
Watching signals and micro-events catch hype windows launches and collabs ignite.
Narrative
Grok analyst briefings
Optional batch narratives translate structured features into language leaders actually read.
[ Operating rhythm ]
How the layer runs
ETL rebuilds `lm_*` profiles per organization, country and marketplace scope.
Paid orders append `lm_events` so windows stay honest between batch runs.
Submit/collect crons turn JSON features into Grok-authored summaries at scale.
Embedding writes power semantic retrieval beside classical ML—same data spine.
Etrovvo
Operate with evidence
Create an organization, connect payments and let Etrovvo learn from real baskets.