Railway Control Simulator with AI Fault Advisor

The project described by the caller is a railway control simulator with an AI fault advisor, built as a hackathon submission for the OpenAI 2026 hackathon. The author states that it simulates railway operations including track circuits, signals, point machines and train movements, and includes an AI component that advises dispatchers and crews on faults and writes incident reports after recovery. The AI is designed to never touch controls, only advise — a safety boundary by design.

The description indicates this is a single-person project with no external team or funding mentioned. It was built using Python and PySide6, with an OpenAI-compatible LLM backend (gpt-5.6-terra), and uses a two-agent development workflow involving Claude Code and Codex CLI. The author emphasizes that the AI does not execute commands, and that deterministic replay is preserved by keeping AI messages separate from domain state.

What changed: The project appears to be a proof-of-concept or prototype built in a short timeframe (a hackathon), with no evidence of commercial traction, revenue, or customer adoption. It is described as educational and safety-focused, not production-ready.

Most important open question: Is this an educational tool for railway training, or does it have potential to evolve into a safety-critical operational system? The description states the AI only advises, but there is no evidence of how this would scale or integrate with real-world systems beyond the simulator.

The description states that the project is:

• A railway control simulator for a virtual single station, including:
  • Track diagrams with 12 track circuits, 4 point machines, 6 signals, and 8 routes.
  • Support for route setting, occupancy, fault indications, fail-safe interlocking.
  • Automatic train operations on a timetable, recording stops, departures, and delays.

• An AI Fault Advisor that:
  • Reads an immutable snapshot of the station state when a fault occurs.
  • Generates dispatcher notifications, crew cautions, and recommended actions.
  • Writes incident reports after fault recovery, including timeline, affected trains, estimated cause, and suggestions to prevent recurrence.

• Built with:
  • A pure-Python domain model separated from UI (PySide6).
  • An OpenAI-compatible client, defaulting to gpt-5.6-terra.
  • A two-agent development workflow using Claude Code and Codex CLI.

The author emphasizes that the AI never executes commands, only advises, and that AI messages are kept separate from domain state to preserve deterministic replay.

The description states:

• The project is an educational railway control simulator.
• It was built to reduce human error in fault response, not replace human judgment.
• The AI advisor helps with drafting notifications and incident reports, based on real-time state data.
• A key design principle is that the AI must not operate equipment — this is a safety boundary by design.

The author frames the project as a test of AI boundaries in safety-critical domains, where the AI’s role is advisory only. There is no claim about commercial deployment, production readiness, or integration with live systems.

This positioning is self-reported and unverified. The project is described as a hackathon prototype, not a product or service.

The description states:

• The author works as a field technician on the railway.
• The simulator is designed to help dispatchers and field staff in fault response.
• It is intended for educational use, possibly training personnel in railway operations and incident reporting.

There is no evidence of:

• Specific customer segments beyond the author’s own role.
• Any commercial or institutional users.
• A defined ICP (Ideal Customer Profile).

The project appears to be aimed at railway professionals or training institutions, but no explicit target customer list or persona is provided.

The description states:

• The project is a hackathon submission, not a commercial product.
• There is no mention of pricing, licensing, or monetization strategies.
• No evidence of revenue, customers, or sales channels.

Not evidenced: Any business model or pricing structure.

The description states:

• Built with:
  • Pure-Python domain model separated from UI (PySide6).
  • An OpenAI-compatible client, defaulting to gpt-5.6-terra.
  • A two-agent development workflow using Claude Code and Codex CLI.

• Key technical features:
  • Fail-safe interlocking for route setting, occupancy, and fault indications.
  • Immutable snapshots for AI input.
  • Non-blocking LLM generation, handled by worker threads.
  • Error isolation: AI failures do not affect simulation or replay.
  • Deterministic replay preserved by separating AI messages from domain state.

• Challenges overcome:
  • Maintaining determinism despite non-deterministic LLM responses.
  • Non-blocking generation during LLM calls.
  • Fixing a design gap in departure handling.

The author also mentions:

• 85 regression tests passing at submission.
• Development workflow halved feature development time.
• Use of gpt-5.6-terra for both build-time and runtime AI tasks.

The description states:

• This is a hackathon project submitted to the OpenAI 2026 hackathon.
• No evidence of:
  • Revenue, customers, or adoption.
  • Product-market fit or user feedback.
  • Post-submission development or deployment.

There is no evidence of traction beyond the author’s own account. The project is described as a proof-of-concept, not a product in use.

The description states:

• No mention of competitors or market analysis.
• The author does not reference existing railway simulators, AI fault systems, or safety-critical control platforms.

Not evidenced: Any competitive landscape or positioning relative to other tools in the railway or AI domains.

• Unverified claims: All features and design decisions are self-reported.
• No commercial traction: The project is a hackathon submission, with no evidence of adoption or revenue.
• Single-person development: No team, funding, or external support mentioned — raises questions about scalability or long-term viability.
• Prototype nature: The author explicitly states this is a proof-of-concept, not a production-ready system.
• AI safety boundary: While the AI is designed to not operate controls, there is no evidence of how this would be enforced in real-world systems beyond simulation.

1. What are the specific use cases for this simulator beyond training?
2. How does the AI’s advisory role translate into a real-world safety-critical system?
3. Is there any plan to integrate with live railway systems, or is it purely educational?
4. What would be the next steps in terms of development, testing, and deployment?
5. How do you plan to scale beyond a single station or extend functionality?
6. Are there any real-world fault case studies or data sources used for training the AI?
7. What is the long-term vision for this project — educational tool, commercial product, or research prototype?