Uploaded November 2025 | Updated September 2026, 2 weeks ago
Accurate geolocation of network infrastructure remains a persistent operational and research challenge, with misplaced routers and transit hops impacting latency tuning, SLA verification, outage forensics, and regulatory compliance. While prior work has shown the value of extracting geographic hints from DNS PTR records, current approaches cannot scale to the diversity of naming schemes used by thousands of ASes.
In this talk, we present The Aleph: an approach that leverages Large Language Models (LLMs) to automatically classify PTR naming patterns, generate extraction regexes, and map extracted hints to real-world locations. Our method operates at Internet scale, covering billions (99.99%) of PTR records from 20k+ ASes.
Our results show substantial gains in coverage and accuracy over existing approaches, with ground truth information from 3 network operators and active measurement validation for 200M+ IPs, giving a 94% confidence in IPv4 and 90% in IPv6. We discuss integration into IPinfo’s production geolocation pipeline, its role in enhancing traceroute interpretation, and opportunities for community adoption through open APIs and datasets.
Kedar Thiagarajan: I’m a third-year PhD student in Computer Science at Northwestern University, where I’m advised by Dr. Fabian Bustamante. My research focuses on Internet measurement and the use of machine learning to tackle networking problems, with an emphasis on understanding the resilience, security, and performance of Internet infrastructure. Before starting my PhD, I worked as a software engineer at VMware and Meta, where I built and scaled distributed systems and production tooling. I received my undergraduate degree in Computer Science from UCLA, where I first developed an interest in Internet infrastructure research.
Accurate geolocation of network infrastructure remains a persistent operational and research challenge, with misplaced routers and transit hops impacting latency tuning, SLA verification, outage forensics, and regulatory compliance. While prior work has shown the value of extracting geographic hints from DNS PTR records, current approaches cannot scale to the diversity of naming schemes used by thousands of ASes.
In this talk, we present The Aleph: an approach that leverages Large Language Models (LLMs) to automatically classify PTR naming patterns, generate extraction regexes, and map extracted hints to real-world locations. Our method operates at Internet scale, covering billions (99.99%) of PTR records from 20k+ ASes.
Our results show substantial gains in coverage and accuracy over existing approaches, with ground truth information from 3 network operators and active measurement validation for 200M+ IPs, giving a 94% confidence in IPv4 and 90% in IPv6. We discuss integration into IPinfo’s production geolocation pipeline, its role in enhancing traceroute interpretation, and opportunities for community adoption through open APIs and datasets.
Kedar Thiagarajan: I’m a third-year PhD student in Computer Science at Northwestern University, where I’m advised by Dr. Fabian Bustamante. My research focuses on Internet measurement and the use of machine learning to tackle networking problems, with an emphasis on understanding the resilience, security, and performance of Internet infrastructure. Before starting my PhD, I worked as a software engineer at VMware and Meta, where I built and scaled distributed systems and production tooling. I received my undergraduate degree in Computer Science from UCLA, where I first developed an interest in Internet infrastructure research.
