JevsBistro
Deterministic 3D restaurant simulator that replays the same evening to compare rule-based, camera-assisted and Jev-planned waiters.
Overview
JevsBistro is a restaurant service simulator. Guests arrive, order, eat, ask for things, pay and leave, while servers carry plates, refill water and clear tables. Everything is deterministic from a scenario seed, so the same evening can be replayed with a different 'brain' driving the servers and compared fairly.
It separates two upgrades: better information (a full-room camera) and better judgment (a fast decision model choosing the next tour of stops). Three conditions are compared: patrol-and-observe rules, camera-assisted rules, and a camera-informed earpiece where Jev (or an OpenAI stand-in) picks the next task. The decision maker only sees timestamped observations, what the server carries and a short list of tasks that are actually possible; any answer naming a task not offered is rejected and the simulation pauses rather than falling back to rules. The simulated clock stops while a request is out, so runs compare decision quality, not latency. Each decision's state, options, confidence and latency are logged.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which of the currently possible tasks should this server do next? | — |
How it works
- Pick a scenario seed and restaurant configuration
- Simulation runs the evening and pauses at each server decision
- The decision maker receives observations and the list of feasible tasks
- Jev picks a task; invalid answers pause the run
- Results are compared across rule, camera and Jev conditions
Strengths
- Deterministic replays enable fair comparisons
- Separates the value of information from the value of judgment
- Logs state, options, confidence and latency per decision
Caveats
- Simulation, not real restaurant operations
- Latency excluded from service outcomes by design
- Requires a Jev or OpenAI key for model conditions
At a glance
- Type
- Demo
- Status
- prototype
- Language
- TypeScript
- Runtime
- browser 3D simulation
- Has code
- yes
Use cases and integrations
How curated lists describe it
Deterministic 3D restaurant sim that replays the same dinner service to compare rule-based, camera-assisted, and Jev-planned waiters, logging each decision's state, options, confidence, and latency.
AnotiaWang/awesome-decision-models
Frequently asked questions
What does Jev decide?
Which feasible task a server should do next, given camera observations.
Why stop the clock during requests?
So the comparison measures decision quality rather than network latency.
What happens if the model names an invalid task?
The answer is rejected and the simulation pauses instead of silently using rules.
Related projects
Browser 2D driving sim where Jev answers lane, speed, hazard and pedestrian-yield questions every 200 ms, with an LLM comparison mode.
ProjectprototypeJavaScriptExperimental Conway's Life game where Jev makes each move with one boolean per legal cell, plus a broker that wraps any LLM as Jev.
ProjectprototypeTypeScriptBrowser lane-racing demo where Jev, OpenAI Decisions and a local EmbeddingGemma model each choose a lane for the same obstacles.
ProjectprototypeJavaScriptEvaluates Jev as a fixed policy in Gymnasium and Atari games, picking a legal action each step via one Choice question.
ProjectresearchPythonModel Kombat: a 12,000-neuron fly connectome fights Jev in mk.js, with Jev picking one of seven moves about five times a second.
ProjectresearchPythonFixed-wing landing simulator and benchmark where Jev maps four Choice decisions to aircraft controls against humans, baselines and LLMs.
DemoprototypePython