Fly the LUS-222 manually.
Take direct control of throttle, pitch and roll with real-time PFD instruments, twin-engine telemetry and optional envelope protection.
PIXELGRAMMAR ENGINEERING USE CASE TYPESAFE-AI/JEV × LUS–222
Take control.
See how typed
AI decides.
A real-time 3D flight simulator engineered by Pixelgrammar to demonstrate a new frontier AI paradigm: JEV (typesafe-ai/jev). Instead of generating chat prose, JEV evaluates structured aircraft telemetry and returns typed decisions, calibrated probabilities and millisecond latency.
ENGINEERED BYPixelgrammar StudioApplied AI · WebGL · Critical UI
DECISION ENGINEtypesafe-ai/jevChoice · Score · Boolean · ~220 ms
TESTBED AIRFRAMELUS–222 Regional STOLPorto LPPR · Human ⇄ AI Handover
Runs in your browser · WebGL ACESFilmic · Instant human takeover at any time
01 / LIVE JEV TELEMETRY & AIRFRAME
Structured flight state in.
Typed decision out.
- MODEL
- typesafe-ai/jev
- PRIMITIVE
- experimental_evaluate
- CONTINUE
- 0.84
- HOLD / GA
- 0.12 / 0.04
- LATENCY
- 214 ms · ZERO PROSE
01 / THE PIXELGRAMMAR USE CASE
Real-time simulation.
Built to test AI under pressure.
Pixelgrammar designs and engineers interactive 3D systems, mission-critical interfaces and applied AI architectures. In this use case, we built a full WebGL flight deck over Porto where a human pilot and TypeSafe AI’s JEV share the exact same aircraft physics, instruments and safety supervisor.
Execute the LPPR approach.
Start on final over Porto, align with the runway threshold, manage sink rate and inspect touchdown telemetry afterwards.
Put JEV in command.
Assign JEV a route or landing objective. Watch its typed choices, calibrated confidence and millisecond latency live—and take over anytime.
02 / WHAT IS JEV & WHY IT MATTERS
No chat prose.
Typed decisions in milliseconds.
JEV (typesafe-ai/jev) is a frontier evaluation model accessed via experimental_evaluate. Unlike conversational LLMs that stream free-form text, JEV ingests structured telemetry and returns schema-locked Choice, Score and Boolean decisions with calibrated probability distributions in a single fast call.
landing-contract.js{
"objective": "land_at_porto",
"phase": "descending",
"altitude_m": 380,
"speed_mps": 59.4,
"vertical_speed_mps": -3.2,
"stable": true,
"engines_available": 2,
"weather": { "tempo": "limpo", "vento_kt": 14 },
"traffic": "clear"
}
experimental_evaluate
typesafe-ai/jev
- Zero free-text generation
- Single-call multi-criteria
- Calibrated probabilities
- ~180–420 ms round-trip
Choice · landingDecisioncontinue 84%
continue 0.84 · hold 0.12 · go_around 0.04
Score · approachRisk18 / 100 LOW
Continuous calibrated rubric [0–100]
Boolean · stabilizedGatetrue CONF 0.97
Verified against descent envelope criteria
Validated telemetry.
Position, altitude, airspeed, wind shear, engine state and nearby hazards are sanitised by strict client contracts before evaluation.
Calibrated choice.
JEV evaluates all criteria in one call and returns a probability distribution over valid actions—clearly distinguishing live AI from recorded replay.
Supervised actuation.
Flight guidance executes the chosen manoeuvre while an independent safety supervisor enforces terrain and traffic separation.
PIXELGRAMMAR CAPABILITIES // ENGINEERING THE STACK
Who is Pixelgrammar—and why we built this.
Pixelgrammar is a software, AI and simulation studio. We partner with ambitious teams to turn new AI models and complex domain physics into high-craft, production-ready interactive products.
Frontier AI Integration
Adapting new model primitives like typesafe-ai/jev into deterministic, schema-validated decision pipelines with real-time observability.
3D & WebGL Simulation
Engineering custom Three.js worlds, procedural terrain, shaders and flight dynamics that run at 60 FPS directly in the browser.
Tactical & Critical UI
Designing high-density instrumentation, glass-cockpit HUDs and human-in-the-loop override controls where clarity is non-negotiable.
GO FURTHER // MULTI-CRITERIA BRIEFINGS
Clinical urgency. Heavy payload. Uncertain sighting.
Stress-test JEV across four operational mission scenarios.
03 / FROM THE LAB
Ideas become clearer
when you can try them.
Pixelgrammar Lab is where we turn questions about AI, interaction and simulation into working experiences. This experiment explores a simple one: what changes when a human and an AI share the controls?
Discover Pixelgrammar (opens in a new tab)