Edge NLP Pipeline for Disaster Summarization

PhD research — offline, edge-deployable NLP system for real-time situation summarization during disasters using social media data.

Overview

This is my primary PhD research project at Symbiosis Institute of Technology (SIT), Pune. The goal is to build edge-deployable NLP pipelines that generate real-time situation summaries from social media data during disaster events — designed to work entirely offline, without cloud dependency.

Current disaster response systems rely heavily on cloud infrastructure, which is often unavailable during the very events they are meant to support. This project addresses that gap by pushing NLP inference to the edge.

Problem Statement

During large-scale disasters (floods, earthquakes, pandemics), first responders need accurate, real-time situational awareness from social media streams. Existing summarization systems:

  • Require internet/cloud access
  • Have high latency for real-time scenarios
  • Are not optimized for resource-constrained edge hardware

Approach

  • Abstractive Summarization of disaster-related social media posts
  • Retrieval-Augmented Generation (RAG) to ground summaries in verified facts
  • Model compression (quantization, distillation) for edge deployment
  • Offline-first architecture — inference runs entirely on-device

Research Interests Addressed

  • Natural Language Processing
  • Crisis Informatics
  • Edge AI & Resource-Constrained Computing
  • Abstractive Summarization
  • Retrieval-Augmented Generation

Status

Ongoing — PhD research at Symbiosis Institute of Technology, Pune (2023–present).