Computer vision · Venue operations
HaKit
Point a camera at a room. It works out what breaks when the plan changes.
- Type
- Computer vision · Venue operations
- Role
- [Add role]
- Year
- 2026
- Status
- Prototype
(01)What it is
HaKit plans the physical side of an event. A camera feed runs through object detection in the browser, and what it sees becomes part of a world state — floors, rooms, exits, obstacles, registrations, check-ins. The app flags issues, proposes plans, and answers “what if?” before it happens.
- Built at
- On my own time
- First commit
- Aug 2026
- File
- 09 of 18 project pages
(02)How it works
- 01→ScanCamera → object detection
- 02→World stateRooms, exits, obstacles, people
- 03→AnalyseIssues against requirements
- 04→What ifConsequence engine
- 05■PlanCandidate arrangements
(03)What's inside
- 01
Vision in the browser
MediaPipe's EfficientDet-Lite2 object detector runs on the GPU in video mode — no server round trip for every frame.
- 02
Every number knows where it came from
Each value in the world state carries its provenance — entered by a person, observed, estimated, simulated, calibrated — and “unknown” is a real value, not a guess.
- 03
Consequences, deterministically
Type “what if 50 more participants” and the consequence engine clones the state, applies the change, re-runs the analysis and shows which issues appear or disappear.
(04)The honest bit
- A prototype: the what-if parser understands a handful of phrasings.
- The venue model is only as good as what the camera and the organiser tell it.
No fake users, no invented numbers. If it's a demo, it says so.
(05)Built with
- App
- React/TypeScript/Vite/React Router
- Vision
- MediaPipe Tasks Vision/EfficientDet-Lite2
- Server
- Express
(06)The world
The same picture the film runs — HaKit drawn from what it actually does, never a screenshot.
(07)Links
- Live[No live demo yet]
- CodeView the code ↗