PulseOps

Inspiration

Transit agencies deal with enormous complexity, yet their tools are overwhelmingly reactive. Delays compound before anyone notices, communication lags, and riders feel the impact instantly. We wanted to build something that gives agencies superpowers—a system that can understand what's happening, predict what’s about to happen, and recommend the best actions in real time. PulseOps was inspired by the idea of bringing AI‑driven operational intelligence to public transit, the same way copilots are transforming other industries.

What it does

PulseOps is an AI-powered transit operations copilot that ingests real‑time GTFS‑RT data, event and weather context, historical disruptions, and learned patterns to create a unified operational command center. It forecasts emerging delays, generates early‑warning risk signals, auto-detects incidents, and produces complete action plans—including rider alerts, operational playbooks, shuttle recommendations, and social posts. It also continuously learns from real-world case studies and operator feedback. The dashboard provides a clean, predictive, and always‑updating picture of how the system is performing right now.

How we built it

We built a full-stack platform powered by TypeScript, Node, and a modular real-time ingestion engine. We integrated MBTA’s GTFS‑RT feeds and Parallel’s FindAll API to pull in real-world disruption case studies. We created a time-series health engine, risk predictor, and a “Transit Brain” knowledge graph that understands corridor vulnerabilities and historical patterns. The UI is a modern, responsive, single‑page dashboard that communicates with structured backend endpoints—including /health, /risk, /context, /brain/insights, /incidents, and /simulate. Claude serves as the reasoning engine for planning and analysis, generating structured response packages for operations.

Challenges we ran into

Handling real-time transit data was more complex than expected—feeds can be sparse, delayed, or inconsistent. Designing thresholds that detect meaningful incidents without flooding the system was also difficult. Integrating multiple intelligence sources (GTFS‑RT, weather, events, case studies) in a way that feels coherent took careful architectural planning. Ensuring the AI planner produced grounded, consistent action plans required multiple iterations. Building a polished UI that gracefully handles empty, loading, and failure states was also a significant challenge.

Accomplishments that we're proud of

We created a fully functional, predictive operations platform in a short timeframe. We built a clean UI, a multi‑source intelligence backend, and an advanced AI planner that produces structured operational responses. We implemented real-time health scoring, risk forecasting, scenario simulation, and real-world case-study matching—all working end-to-end. The system feels like a real product, not just a hackathon prototype. And it genuinely improves how transit agencies can understand and react to disruptions.

What we learned

We learned how difficult but impactful it is to merge historical intelligence, real-time data, and AI reasoning into a single system. We learned the importance of clean data shaping for LLMs, how to design structured prompts, and how to create human‑interpretable operational metrics. We also discovered how effective knowledge graphs and case-based reasoning can be when guiding AI-generated decisions. Most importantly, we learned how AI can elevate public transit operations into something proactive instead of reactive.

What's next for PulseOps

Next, we want to integrate deeper crowding estimation, richer event impact modeling, and more granular headway and capacity forecasts. We plan to expand our Transit Brain knowledge graph to support more agencies and modes. We’ll add automated outbound communications (Slack, SMS, GTFS‑RT alert generation) and bring in reinforcement learning loops that adapt thresholds and playbooks automatically. Long-term, PulseOps can evolve into a fully autonomous AI dispatch assistant that predicts disruptions, assigns resources, and optimizes service reliability across entire systems.

Built With

• alerts)
• case-studies
• claude-(anthropic-api)
• css
• custom-responsive-css-ui
• express?style-routing
• fetch-api
• gtfs?rt-(mbta-trip-updates
• html
• javascript
• json?based-local-storage-(learning-log
• knowledge-graph)
• node.js
• optional-external-weather-and-events-apis
• parallel-findall-api
• rest
• typescript
• vanilla-js-single?page-app
• vehicle-positions