TranistPulse

Inspiration

TransitPulse was inspired by a simple observation: everyday infrastructure problems can remain unnoticed, unresolved, or poorly prioritized even in large and busy transportation systems. A broken light, malfunctioning display, damaged seating, waterlogging, or cleanliness issue may appear small individually, but collectively they affect safety, accessibility, reliability, and passenger experience.

I wanted to explore whether AI could help bridge the gap between noticing an infrastructure problem and turning that observation into useful maintenance intelligence.

What it does

TransitPulse is an AI-powered infrastructure intelligence platform that transforms real-world observations into structured, actionable maintenance insights.

A user can submit an observation with a description, location, and visual evidence. The system uses AI to interpret the information, classify the issue, assess its preliminary severity and impact, and suggest an appropriate maintenance action.

The goal is not to replace existing maintenance systems or human decision-makers. Instead, TransitPulse is designed as an intelligence layer that can help organize, prioritize, and contextualize infrastructure issues before they reach the people responsible for resolving them.

How I built it

I built TransitPulse as a lightweight web application with a simple architecture so that the core workflow remains easy to understand and demonstrate.

The frontend provides the interface for submitting infrastructure observations and viewing maintenance information. A FastAPI backend handles the application logic and communication between the interface and AI services. Gemini provides multimodal AI capabilities for interpreting descriptions and visual evidence, while AWS provides the cloud infrastructure and services needed to support the application.

I deliberately kept the architecture simple rather than introducing unnecessary microservices, complex infrastructure, or custom AI models. The central workflow is:

Observation → AI interpretation → Structured incident → Prioritization → Maintenance action

Challenges I ran into

One of the biggest challenges was defining what the actual problem should be. Railway and infrastructure maintenance already has sophisticated enterprise solutions, so I did not want to claim that existing technologies cannot solve the problem.

I instead focused on a narrower gap: transforming unstructured, everyday observations into structured and actionable maintenance intelligence.

Another challenge was deciding where AI genuinely adds value. A general-purpose AI model can already analyze an image or summarize a complaint, so simply putting an AI chatbot on top of a reporting form would not be meaningful. I therefore designed TransitPulse around the workflow surrounding the AI: classification, contextualization, prioritization, and maintenance tracking.

Keeping the prototype simple while still demonstrating this complete workflow was another important constraint.

Accomplishments that I'm proud of

I'm proud that TransitPulse focuses on a practical problem rather than treating AI as the product itself.

I was able to define a clear primitive for the system:

Unstructured observation → Structured maintenance intelligence

I also designed the prototype so that AI remains one component of a larger human-in-the-loop workflow. This makes the concept more realistic for organizations that already have maintenance teams and operational systems.

What I learned

I learned that identifying a problem is only the beginning of building a useful AI product. Existing technologies need to be studied carefully before claiming that a new solution is necessary.

I also learned that the strongest use of AI is not always replacing an existing process. Sometimes its value comes from reducing the amount of manual interpretation required between an observation and an informed decision.

Most importantly, I learned to think about AI as part of an operational workflow rather than as a standalone chatbot.

What's next for TransitPulse

The current prototype is intentionally lightweight. The next step would be to connect TransitPulse more deeply with real maintenance workflows and existing asset-management systems.

Future versions could support automated grouping of related reports, recurring-issue detection, historical trend analysis, richer location intelligence, integration with sensors and inspection systems, and more advanced prioritization.

Ultimately, I want TransitPulse to become a practical intelligence layer that helps organizations move from “someone noticed a problem” to “the right problem was identified, prioritized, and acted upon.”

Built With

• amazon-web-services
• api
• css
• fastapi
• gemini
• html
• javascript
• python
• rest
• tailwind