Pilot Proposal · May 2026

TransitSafe Maritime

Pre-survey AI inspection. Crew or technical managers grade their own vessel in 30 minutes — against the same Q-IDs the official surveyor will cite — so defects are found, fixed, and audit-trailed before the surveyor arrives.

A SIRE 2.0 vetting inspection takes 8 hours on board and £4–8k per vessel. Detained findings cost £10k+/day in downtime. A manager's pre-survey self-check with TransitSafe costs nothing, runs continuously, and lands defects in port — not in front of the surveyor.

Why pre-survey wins

The high-value buyer is the operator who pays the cost of a failed inspection — not the inspector who finds it. Same photographic evidence, multiple regulatory corpora, multiple buyers in the chain.

TodayWith TransitSafe pre-survey
Surveyor arrives → finds defects → vessel detained → £10k+/day downtime Crew checks weekly → defects flagged 2–6 weeks before surveyor → fixed in port → surveyor finds clean vessel
Photo documentation lives in spreadsheets, OneDrive folders Same photos auto-graded, cited to Q-ID, archived with corpus version
"Did anyone check the manifold drip tray this quarter?" → file search Continuous condition view by zone, by Q-ID, by date
Audit trail starts with the surveyor Audit trail starts with the crew
Inspector cost per vessel: £4,000–8,000 Same inspector, but they arrive at a clean vessel — fewer findings, faster turnaround

Who buys this

Four buyer personas across the inspection chain:

How the audit trail works

Every photo finding in our report carries three identifiers an underwriter or class society can verify:

  1. SIRE 2.0 Q-ID — e.g. Q11.1.15 (Starboard manifold). Maps to OCIMF Question Library Part 2, page 537, Chapter 11.
  2. Corpus claim ID — e.g. sire2-q11-1-15-starboard-manifold. Internal handle, testable, versioned, license-tracked.
  3. Corpus version — e.g. v0.9.0. SemVer-tracked; same vessel inspected at v0.9.0 and v1.0.0 produces comparable findings against a stable Q-ID — different opinions, same questions.

Sample finding from today's demo:

Cargo Systems — Q11.1.15 (Starboard manifold) Fail · Critical

"Drip tray overflow with visible product residue and a torn vapor-recovery vent screen on the upstream side. Active hazard per Chapter 11 §11.1.1 negative-observation grounds."

Remediation: Vessel out of service for cargo operations until drip tray cleared, vent screen replaced, and pressure test passed.

SIRE 2.0 Q11.1.15 · claim sire2-q11-1-15-starboard-manifold · corpus v0.9.0

Pilot architecture

Crew walkaround (phone / tablet) ←─── continuous, weekly │ ▼ inspection-worker (in-worker AI pipeline) ──── multi-LLM consensus │ (Claude / GPT-4o / Gemini) ▼ maritime-code-rag-worker ◄──── Corpora (versioned): │ • SIRE 2.0 Q-Library (Chapter 11 v0.9.0 → v1.0.0) │ • Class society guidance (pending real survey data) │ • P&I + port state (later corpora) ▼ report-worker (TransitSafe template) │ ▼ Pre-survey report — graded zones, cited Q-IDs, audit-ready

Training data is the gate

What you see today (v0.9.0) is built on public SIRE Q-Library text. Production v1.0 needs four things only an industry insider can unlock:

  1. A maritime technical contact — an operator-side champion who'd run a pilot. Identifies the first paying client.
  2. Compliance documentation beyond the public SIRE Q-Library — class-society guidance, P&I requirements, port-state inspection rubrics.
  3. Example survey reports — real findings, real language, real defect taxonomies. This is what trains the model on how surveyors actually word negative observations.
  4. Surveyor network participation — surveyors are the validation loop: did our pre-check agree with the human result?

Roadmap

May 2026
v0.9.0 pre-survey demo (Chapter 11 SIRE, 8 photo claims, multi-LLM)
✅ shipped
Jun–Jul
Industry contacts, compliance docs, sample survey reports, surveyor introductions
Affinity
Q3 2026
v1.0.0 corpus ingestion (real survey reports + extended SIRE)
Gated on Jul deliverables
Q4 2026
First paying pilot vessel inspection (manager pre-survey)
Pilot client
Q1 2027
v1.1.0 (class society + P&I corpora)
Pilot revenue
Q2 2027
v2.0.0 — vessel-type fan-out (LPG / LNG)
2nd client