We're starting with commerce because it's easy to check — a wrong product is obviously wrong, no debate needed. But we're not building a commerce company. We're building the understanding layer, and commerce is just where we prove it works first.
Netflix doesn't own theatres. Uber doesn't own taxis.
Sequena doesn't need to own the store, the inventory, or the menu — it owns the understanding between what someone means and what an application can act on.
This is direction, not a roadmap of products being built today. Commerce remains the entire near-term focus.
The proving wedge — quick commerce, marketplaces, grocery, fashion.
Dish, cuisine and preference-aware ordering and discovery.
Voice-to-business-action for the shop floor and back office.
Structured procurement intent — quantities, specs, compliance, budget.
Natural-language trip, fare and preference discovery.
"Play a sad Telugu song from a Mahesh Babu film" is how people actually ask. Spotify and Apple Music need the exact title — mood, language and film context don't survive the search bar.
"Something light in Telugu, under two hours" is a real request. Streaming search still expects a title you already know.
Feeds are driven by behaviour, but the moment someone types a search — in their own language, describing a vibe — the same gap reappears.
Vehicle, model and part-compatibility intent for auto parts and service.
To be clear: commerce is the only vertical we're building today. The rest are listed to show where the same intent layer applies — one shared schema and evaluation framework underneath, not a separate model per domain. None of them are active workstreams.
We'd rather tell you plainly which of these is real today than let you assume the wrong one.
Synthetic data is never treated as benchmark evidence. A held-out test set stays protected from training.
Three things changed at once, and they only recently overlapped.
India's newest internet users don't type in English. They speak Telugu, Hindi, Tamil — into search bars built for English keywords.
Ten-minute delivery moved past metros. The catalogs followed, but the search didn't — it's still indexed the way it was for Bengaluru.
Small open models now handle code-mixed Indian language well enough to sit in a live search path — economically impossible a few years ago.
The gap between how India shops and how India's commerce software listens has never been wider — or more fixable.
This is where the gap between how people talk and how software listens is widest — and the hardest place to fake understanding. Prove it here first, then take it everywhere.