What kind of day are we walking into?
Not just weather and bookings. The system should weigh calendar rhythm, local events, hospital pressure, school terms, venue history and expected footfall.
Camera-first AI for real hospitality operations
See the shift before the sales report explains it too late.
SentryX is building a camera-driven operations layer for restaurants and cafés. It turns CCTV, POS, labour data and local context into specific decisions about queues, staffing, table flow, prep, menus and layout.
CCTV and edge processing
FOH and BOH flow
Square and Deputy-ready vision
Operator approves actions
Rain clears at 6: walk-ins recover, but pickup pressure shifts later.
Hospital precinct busy: short dwell lunch traffic rises near the venue.
School holidays: family tables increase while regular office traffic drops.
Comedy show nearby: pre-show dining spike starts before POS trend catches it.
Long weekend Friday: locals leave, tourists hunt brunch later.
Menu item pressure: one popular item delays the pass despite strong sales.
Rain clears at 6: walk-ins recover, but pickup pressure shifts later.
Hospital precinct busy: short dwell lunch traffic rises near the venue.
School holidays: family tables increase while regular office traffic drops.
Comedy show nearby: pre-show dining spike starts before POS trend catches it.
Long weekend Friday: locals leave, tourists hunt brunch later.
Menu item pressure: one popular item delays the pass despite strong sales.
Audio brief
Listen to the SentryX overview.
A short audio overview explaining the restaurant bottleneck problem, the camera-first SentryX concept, and why venue-specific intelligence matters.
The questions a sales report cannot answer.
POS tells you what was sold. SentryX is being designed to explain the physical operating story behind the result.
Where is pressure forming right now?
Queue at ordering, pickup congestion, pass delay, BOH pressure, slow table clearing, delivery-driver clustering or a layout pinch point.
What actually caused the outcome?
Strong sales may still hide avoidable stress. Weak sales may not be weak demand. The goal is to separate the visible cause from the easy story.
Not another AI forecast glued to POS data.
The public benchmark is already moving toward explainable forecasts. SentryX has to go deeper by adding the physical layer: cameras, edge processing and venue-specific ML.
Capability
Forecasting layer
SentryX direction
Revenue forecast
Weather, events, trading history and manager notes.
Forecast plus visible demand pressure and flow.
Live service
Margin warnings and generic suggestions.
Camera-derived queue, table, pass and congestion metrics.
Recommendation
Plain-English explanation of the likely number.
Specific action: move staff, open a point, alter prep, change a menu, fix a layout issue.
Automation
Manager still acts manually.
Operator approves, then connected systems can update where permitted.
The kind of signals it should catch.
These examples are deliberately concrete. The point is not a generic dashboard. The point is a system that remembers what a venue feels like when the numbers move.
Rain ends before dinner.
Daily sales look soft, but camera flow shows walk-ins recovering after 6. The system warns against cutting too early.
A sign creates a pinch point.
Customers cluster near pickup while tables remain clear. The issue is not demand. It is physical friction.
The wrong person is in the wrong zone.
Comparable rushes show one staff member performs better at handover than till. The system suggests a roster split change.
Hospital activity lifts lunch traffic.
A nearby health precinct is under pressure. Short dwell visitors rise, and the model learns when that signal matters.
A popular item slows the pass.
POS says it sells well. Camera timing shows it creates pickup congestion. The recommendation is prep, menu or availability change.
Locals leave, tourists arrive later.
A long weekend does not just reduce regulars. It changes table dwell, ordering time and the brunch window.
What we are building toward.
The first product is decision support. The mature product is operator-approved execution across the venue’s existing systems.
Not the goal
- Not a generic chatbot sitting on top of POS exports.
- Not raw camera footage sent to a central server by default.
- Not an identity database or staff punishment tool.
- Not vague summaries like “rain affected trade”.
The actual goal
- Camera-first operational metrics from real venue flow.
- Local edge processing for sensitive visual signals.
- Venue-specific ML that learns from operator feedback.
- Approved actions pushed to Square, Deputy or menu systems where integration allows.
Design principles.
Camera-led systems need clear boundaries. The product should feel useful to operators, not creepy to staff or customers.
1
Camera-first, not camera-only
Cameras provide the physical operating truth. POS, labour and external data explain the context around it.
2
Local by default
Raw or identifiable visual data should stay on venue hardware where practical. Central systems receive filtered outputs.
3
Explainable enough to trust
Recommendations should show the signals behind them. Operators can approve, reject or correct the system.
4
Specific action over generic insight
Move this person. Open this point. Change this prep. Hide this item for 30 minutes. Fix this layout friction.
How a pilot would work.
The first venue does not need a finished product. It needs operational richness, usable cameras and a willingness to test whether the system sees what managers already sense.
Map the floor
Service zones, camera angles, queue paths, table areas, pickup points, pass flow and staff movement.
Pick one painful metric
Queue pressure, table dwell, pass congestion, service-zone blockage or roster mismatch.
Connect context
POS, Square, Deputy, weather, holidays, nearby events, bookings and operator notes where available.
Judge the output
Compare system recommendations with operator judgement, then decide whether to expand venue by venue.
Good first fit, poor first fit.
SentryX is not meant for every venue on day one. It needs repeatable pressure and enough activity to learn from.
Good pilot fit
- Busy independent restaurant, café or quick-service venue.
- Clear rushes, recurring bottlenecks or service pressure.
- Existing CCTV or openness to camera mapping.
- Square, Lightspeed, Deputy or similar systems already in use.
- Operator willing to give honest feedback on the recommendations.
Poor first fit
- Very low-volume venue with limited repeatable patterns.
- No useful camera coverage and no appetite to test it.
- Expectation that AI instantly solves management.
- Use case focused on monitoring individuals instead of improving operations.
- No willingness to compare output with real venue judgement.
Early questions.
These are the questions operators, investors and pilot partners are likely to ask first.
Is this already a finished product?
No. SentryX is in concept and proof-layer planning. The immediate goal is to build a pilot-ready camera-first system that can be tested against real venue judgement.
Does this replace a manager?
No. The operator remains in control. The system should recommend, explain, ask for approval and learn from overrides. Later, approved actions can flow into connected systems.
What data would it use?
CCTV/camera feeds, queue patterns, table dwell, staff/customer flow, POS, labour, Square-style menus, Deputy-style rosters, weather, holidays, events, hospital precinct signals and operator notes.
Why camera-first?
Hospitality is physical. POS cannot tell whether the problem was weak demand, a blocked pickup area, poor table flow, an overloaded pass or the wrong staff split.
How does privacy work?
The preferred architecture is local edge processing. Raw or identifiable visual data stays local by default. Central systems receive non-identifying outputs, aggregate metrics or model feedback where permitted.
Who is behind it.
A small technical and operations-led team shaping the product around real venue pressure, not generic AI hype.
Rishabh Desai
Founder · Product · Systems · Operations
Leads product direction, hospitality problem definition, systems framing, customer discovery, operational logic and day-to-day execution.
Pratham Shenwai
Co-founder · Technical ML Lead
Supports machine learning development, computer vision thinking, model design, data pipelines, signal filtering and implementation planning.
Building toward a serious pilot.
SentryX is seeking practical conversations with venue operators, pilot partners and early-stage supporters who understand that hospitality intelligence has to be built from the floor up.