Products
The Barvis line
Every bar knows what it sold. Almost none know what actually left the shelf. Barvis puts one camera over the pour zone and produces an independent count of every drink poured, then reconciles that count against your POS. Where the two records disagree, that is revenue leaving without a ticket. All processing happens on a box in your venue — no video leaves the premises.
Why a second record
You know what you sold. You do not know what left the shelf. The difference is money, and right now nobody is counting it.
01
The gap is real and it is yours
A 750 ml bottle should yield 12.5 pegs. Most operators do not know what theirs actually yields, because nothing in the room measures it.
02
A stock count cannot fix it
It tells you something went missing. It cannot tell you when, at which station, or on which SKU — so nobody acts on it.
03
A second, independent record can
A camera over the pour zone produces its own count, and two records that disagree are information a single record can never give you.
04
It is honest about what it can measure
Peg granularity, confidence bands, reconciliation over windows. We publish our error floor, which is not something you will hear from the next vendor.
05
It is safe to put in your room
No facial recognition. No video leaving the building. Attribution to a station, never to a person.
Barvis-X1 Vision
The flagship model. A camera over the pour zone produces its own count of every drink poured — at peg granularity, with a confidence band.
Barvis-X1 watches the pour zone and nothing else. It detects each pour, attributes it to a bottle and a station, and emits a count with a confidence band. It does not guess millilitres from pixels, because free-pour volume cannot honestly be recovered at that resolution.
What you get is a second record of every drink that left the bottle. No facial identification — the model is trained on the pour, not on the person.
Full specificationReconciliation Engine
Matches the vision ledger against the POS ledger and surfaces the divergence. Source-agnostic by design.
Two independent records of the same event, aligned by time, station and SKU. Where they agree, nothing is reported. Where they diverge, that difference is the entire product.
The engine reports variance rather than blame — sustained negative variance flags leakage, positive variance flags under-pouring or a measurement problem worth fixing. Your POS is one writer into the ledger, not the authority over it, and the matching engine is the durable asset.
Full specificationBarvis Edge Node
One box in the venue. All real-time inference runs on premises; raw video never leaves the building.
A single box, sized for your room, sitting on your power and your network. Every frame is processed inside your building. What crosses the network is a stream of counts — timestamp, station, SKU, pegs, confidence — orders of magnitude smaller than video and carrying no imagery.
There is no cloud video pipeline to turn on later. Edge-first is simultaneously the economics and the privacy posture: cloud GPU per venue is unaffordable at the scale this product needs to reach, and a policy you can switch off is not an answer to a camera pointed at people at work.
Full specificationVenue Analytics
The same cameras, already installed and already trusted, extended to queue length, occupancy and footfall.
The expensive part of computer vision in a venue is not the model. It is getting a camera mounted, a box installed, a network agreed, and a room to trust what the system says about it. Once that has happened, asking a second question costs almost nothing.
Venue Analytics is that second question. How long is the queue at the bar, and when. Which zones fill and which stay empty. How footfall moves across the hours of a night, and how that maps onto what was poured. Staffing, layout and opening hours are currently set by instinct and a rough memory of last Saturday; this replaces that with a record.
Planned for 2027
It is not part of what ships today, and the year goes on it every time we mention it.
It runs on infrastructure you will already have
Same edge node, same cameras. The marginal cost of the second question is close to zero — that is the actual argument, and it is the only one we are making.
Same privacy posture
No faces, no individual tracking, no video egress. Occupancy is a density, not a set of people.
The camera you install this year is the camera that answers next year’s questions. That is the whole argument for going deep on the first one.
Distributor Intelligence
Pour-by-pour, SKU-by-SKU data accumulating across venues — the layer where an operational tool becomes a data business.
Below the depot, the Indian liquor market is measured by inference. Dispatch data tells a distributor what left the warehouse. Survey panels tell a brand owner roughly what a city drinks. Between those two is the actual event — a specific SKU, poured in a specific kind of room, at a specific hour of a specific night — and nobody has ever recorded it.
Barvis records it as a by-product of doing something else. Aggregated across a market and anonymised, that becomes something neither dispatch data nor a survey can be: a measurement. This is a later phase, and it depends entirely on earning the operator’s trust first. The operational tool comes first because it has to.
Aggregated and anonymised, always
A venue’s identifiable operational data is never sold. Your data is yours.
Consent is in the agreement
The venue agreement states the aggregated use explicitly. It is not buried and it is not assumed.
Sample honesty
A report says how many venues, in which city, over what period. A market read from eleven bars is a market read from eleven bars, and it says so.
Want to see it run in your venue?