OpenDriveEDU
Clipboard · Drive sheet
WA DOL DTS-661-047 (N/12/24) · AI-drafted · Instructor-reviewed · Licensed instructor signed
Beacon records. Clipboard drafts. The licensed instructor decides.
Clipboard prepares suggested 1–4 ratings for each observed drive-sheet skill row, plus an OpenDrive EDU Signal Panel snapshot, eco score, behavior snapshot, Section 3 comment draft, and run-specific student/family practice suggestions. The licensed instructor reviews, edits, confirms, and signs. No AI output becomes official without instructor approval.
Guided demo (3 min)
A narrated pass through the synthetic drive-sheet workflow, the review gate, and the boundary that keeps every draft under licensed instructor control.
This is OpenDrive Clipboard — our submission for the Google for Startups AI Agents Challenge. One sentence before anything else: Beacon records. Clipboard drafts. The licensed instructor decides.
Behind-the-wheel lessons fail in two ways. They drift into feeling like tests, or they end in a pile of corrections nobody remembers. Clipboard turns one hour of driving into a structured, evidence-informed debrief — drafted by an agent, decided by a licensed instructor.
Here is the judge workflow. Pick a synthetic one-hour drive from the curriculum. Every record on this page is sample data — no student PII, no licensing records, no DOL filings. The counter says it plainly: official records created, zero.
Run the agent and watch the tool chain: scenario intake, context retrieval, drafting, routing — a deterministic, MCP-style chain. Gemini shapes the prose. It never alters a rating, a signal value, or a score. The numbers are reproducible per scenario.
What comes back is a draft of Washington's actual drive sheet — DOL form D-T-S six-six-one, zero-four-seven — with suggested one-to-four ratings, row by row. Two rules hold the line. Rows the scenario never observed are marked NO SIGNAL — the agent does not guess. And evidence never transfers between rows: a stop-sign event cannot support a traffic-light rating.
Alongside the sheet: an eco score, a behavior snapshot, and Section three comments — each labeled synthetic, none of them a clinical assessment or an official measurement.
Then the part that matters most. Every draft dead-ends at the instructor review gate. Status: draft — instructor review required. The licensed instructor reads each row, overrides any rating, then approves, edits, rejects, or regenerates. Signatures unlock a watermarked SAMPLE packet — an audit copy and a student take-home. Not DOL official scoring. Not a legal determination. Not autonomous control.
One more thing. This is not a paper architecture. Behind the demo is a working edge stack — a Jetson reading the CAN bus, read-only, with an onboard context model at sub-millisecond inference, bench-verified across fifty miles of real driving. The public demo runs synthetic data shaped to that real output.
AI organizes the moment. The instructor leads the lesson. That boundary is the product.
OpenDrive Clipboard, from OpenDrive EDU.
Scenario
Each card is a one-hour BTW lesson from the GrowthDrive curriculum. The competence bar shows the deterministic improvement arc the synthetic learner traces — Drive 1 is rough, Drive 6 is test-ready.
Use a fake first name only. The rest of the student and school fields are prefilled SAMPLE data.
Selected: Choose drive number
Numbers are deterministic per scenario. The optional Gemini provider may vary prose only. It never alters suggested 1–4 ratings, OpenDrive EDU Signal Panel values, eco score, or behavior-profile percentages. All outputs remain DRAFT — INSTRUCTOR REVIEW REQUIRED.
Agent workbench
Bounded action · What the agent is allowed to do
Bounded action · What the agent must not do
For Google for Startups AI Agents Challenge judges
Autonomy: bounded, deterministic, observable.
The agent ingests one synthetic Beacon scenario, runs deterministic passes in sequence (telemetry → OpenDrive EDU signal panel → eco → behavior snapshot → skill ratings), then composes Section 3 comments and a fresh student/parent suggestion seeded by the generated report id. Every input and every output is on this page; nothing leaves it.
Human-in-the-loop: the Review Gate.
Below the report you'll find Approve / Edit / Reject / Regenerate. The agent does not act on those decisions — it only records the instructor's choice for the audit trail. The DOL examiner remains the only authority on official scoring.
Boundary · Synthetic data · Instructor-reviewed · No DOL writes
This page drafts a DTS-661-047-style drive-sheet artifact from synthetic scenarios. It is not a licensing, grading, diagnostic, enforcement, vehicle-control, student-facing instruction, or automatic DOL submission tool.
Clipboard does not submit LMS records, certificates, transcripts, licensing records, or DOL records.
The licensed instructor remains responsible for all official teaching, scoring, documentation, signatures, and record decisions. See Boundary and Clipboard demo.