Infra Pulse

• 

  Mapping high-risk road segments

Public infrastructure failures often show warning signals early, but those signals are scattered across inspection notes, complaint logs, and condition metrics. We built InfraPulse to unify those fragmented signals into one operational view that helps teams prioritize risk before incidents escalate.

Features: Calculating per-asset risk and inconsistency scores Surface map hotspots and heightened activity clusters Explaining likely root causes using clustered report language Voice/text inspection-note ingestion and signals updates after new reports Recommendations on practical follow-up actions for field teams

Implementation: Frontend: Next.js + TypeScript + Tailwind Backend: Express API layer for orchestration/proxy, Python logic ML API: FastAPI service for scoring and retrieval ML logic: Term frequency-inverse document frequency embeddings (NLP), cosine similarity, KNN-style nearest incident retrieval, and KMeans clustering. Data: Currently using synthetic offline datasets for roads, bridges, and reports; local artifacts + JSONL feedback persistence. This is highly scalable to real-world incident report data.

Accomplishments: Delivered a working end-to-end MVP with map intelligence, activity clustering, and inspection ingestion Built an explainable risk pipeline instead of a black-box score Added robust fallback behavior so demo UX still works when ML services are unavailable Kept the architecture modular and extensible for future data/model upgrades

Built With

• data-analysis
• express-api
• fastapi
• knn
• machine-learning
• next.js
• tailwind
• typescript