AGENTS, EVERYWHERE · VALENCIA

Fire Incident Copilot

Every claim.
Back to
its source.

Radio traffic moves fast. Keep the evidence, the questions and what’s still unknown in one operator workspace.

BUILT BY JSLAB12 SEPTEMBER 2026HACKATHON PROTOTYPE
Fire Incident Copilot artwork: a radio, an illuminated building model and the prototype operator interface.
AI-generated illustration + prototype interface
ONE WORKSPACE.
MULTIPLE SOURCES.
01 Radio traffic02 Camera evidence03 Sensor readings04 Human judgment

01 / THE QUESTION THAT MATTERS

Was the channel assignment
actually acknowledged?

A request, an assignment and a reply arrive as separate messages. Our focused workflow connects the exchange while keeping each piece of evidence visible.

01SOURCE REPORT

The request.

A group asks for a working channel. Open a check and retain the original source.

What was requested?
02SOURCE REPORT

The assignment.

A channel is assigned. Link it to the request without assuming it was received.

Was an answer given?
03SUPPORTING REPLY

The acknowledgement.

A separate reply acknowledges the channel. Keep the supporting reference beside the result.

What does the reply establish?

02 / INSIDE THE WORKSPACE

Context you can
come back to.

Follow a check as new reports arrive. Ask a question. Inspect the source references behind the answer.

FIREWATCH / INCIDENT COPILOTPROTOTYPE
Actual Firewatch prototype: Incident Copilot question field, three linked sources, an active check, and supporting camera and sensor panels. The visible demo uses FixtureEngine.
ACTUAL PROTOTYPE SCREENSHOT · CLICK TO EXPANDEvidence first. Unknowns visible.
A

Keep the thread.

Background checks maintain an evolving evidence card as the exchange develops.

B

Ask in context.

Questions use a bounded set of source events, with references and stated limitations.

C

Inspect the evidence.

Read timestamps and source reports. The replay interface lets operators revisit recorded radio intervals.

What you’re seeing

A hackathon prototype with a 206-second historical Palisades radio replay and prerecorded machine transcripts. Synthetic Base2 building sensors and CCTV are separately labelled. The verified agent demonstration uses a deterministic test engine; end-to-end live-model analysis remains unverified. It supports a human operator and does not issue tactical instructions.

03 / BUILT FOR THE ENVIRONMENT

The sources shape
the agent.

The assistant sits beside the incident timeline. Source IDs, timestamps, changing evidence and missing information shape its workflow.

OpenAI Agents SDK powers the implemented extraction and question-answering components, with an OpenRouter provider integration. A Python telemetry platform, REST, SSE and MCP connect the components; the browser presents the operator’s working context.

OpenAI Agents SDKOpenRouterPython / FastAPIPostgreSQLSSE / MCPJavaScriptBlenderHiggsfield
Explore the implementation

04 / THE PEOPLE BEHIND IT

Four builders.
One shared context.

JSLab · AI Tinkerers Valencia · September 2026

Anastasiya Nikiforova

Anastasiya Nikiforova

Operator experience & dashboard

Vitalii Nechitailo

Vitalii Nechitailo

Telemetry platform & data APIs

FIRE INCIDENT COPILOT / JSLAB

Keep the evidence.
Keep the context.

See what we built