Making Cities Awesome: Peregrine’s Nick Noone & Ben Rudolph

Summary

Nick Noone and Ben Rudolph describe Peregrine as a privacy-first public-safety infrastructure company built around cities’ existing data rather than new sensor networks or data collection. Their thesis is that safety is the base layer for making cities better, but that trust depends on data governance, permissioning, and preserving each institution’s ownership and control. The founders trace the company’s origins through humanitarian work, early rejection by public-safety agencies, and eventual access at San Pablo PD in 2018. They also argue that most of the hard work happens before a user asks a question: data preparation, integration, semantic structure, and citation accuracy are what make AI useful in high-stakes workflows. The episode highlights agentic systems that reproduced cold-case results, identified weather-driven rescue spikes, surfaced anti-Semitic threat patterns, and even generated custom tools like hurricane simulators and city-planning models.

Notes

• Peregrine frames its mission as helping cities, counties, and states improve the places people live.
• Safety is the company’s foundational civic layer.
• Nick links city progress to both objective safety and perceived safety.
• The founders tie civic infrastructure to preserving individuality and privacy.
• Forward-deployed engineering is defined as owning or co-owning customer problems end to end.
• Nick says the goal is to reach outcomes three to five times faster than others.
• The company treats customer success, not vendor credit, as the win.
• Nick says Silicon Valley underestimates the institutional and human context inside police departments.
• He argues that speed inside institutions is counterintuitive to many customers.
• Ben’s thesis came from humanitarian work and refugee-border deployments.
• He says UNHCR’s fragmentation and spreadsheet-heavy workflows revealed downstream data problems.
• The company’s initial wedge was public safety.
• Nick says they got into San Pablo PD on February 26, 2018 after more than two dozen rejections.
• The early outreach was trust-based and grounded in learning from real investigations.
• Ben says the company’s work is apolitical and aimed at what public institutions actually need.
• Peregrine positions itself as the inverse of collection-first public-safety vendors.
• The platform is designed to sit atop existing systems, not replace them.
• Each institution is supposed to own its own data.
• Peregrine emphasizes selective sharing, not broad data exposure.
• AI deployments often begin as better search.
• As users learn the product, the department-wide baseline rises.
• Ben says the deeper value is analysis previously impossible.
• A Florida county used integrated incident, 911, and weather data to investigate water rescues.
• The model inferred that three-day weather patterns created sand channels and rip currents.
• Semantic AI found anti-Semitic threat patterns that keyword search would miss.
• Ben says AI can semantically understand access-controlled evidence and surface threat patterns.
• Ben says roughly 95% of the work happens before the question is asked.
• The hardest problem is making answers accurate and properly cited.
• Peregrine also enables non-agent custom software like hurricane simulators.
• The deployment team often helps customers build their own applications.
• Half of the engineering team focuses on the data platform.
• Peregrine says it has integrated tens of thousands of datasets.
• Agents now write most Python-notebook integration code under deployment-team oversight.
• The first agentic use case was cold-case analysis.
• That agent handled 200 to 300 gigabytes of mixed evidence.
• The evidence set also included paper records, CDs, and cell call-detail records.
• The agent eventually reproduced detectives’ results.
• Trust is protected by not claiming the customer’s wins as Peregrine’s own.
• Nick says public celebration can break trust with sensitive institutions.
• The company frames its politics as anti-authoritarian and anti-network-effect.
• Data governance and permissioning become central AI design problems.
• Nick argues against vendor-imposed general-purpose policy on facial recognition.
• Instead, Peregrine tries to help customers weigh local context, laws, and norms.
• Most public-safety agencies do not implement facial recognition.
• The platform stack spans access, permissioning, governance, ETL, pipeline, ontology, UX, and APIs.
• Peregrine wants that stack to remain open and interoperable.
• Scaling requires trusting individual contributors in difficult field environments.
• Nick compares deployment to sending a person into a cave with tools and a rope back to base.
• Forward-deployed engineers are expected to maintain integrity and principles.
• Forward-deployed work is treated as R&D and growth.
• Customer pilots are used to sprint on additional use cases after landing.
• A field-editing primitive was initially hacked via comments on objects.
• The hack was a signal that native product primitives were needed.
• Ben treats those hacks as rapid prototyping.
• The company distinguishes core primitives from one-off innovation-lab builds.
• A deployment strategist built a hurricane simulator inside Peregrine.
• Another custom tool estimated population impact from fire-department placement using multiple data sources.
• The long-term end state is serving 10,000 cities.
• Nick says the operating model must be infrastructure-oriented.
• The company wants governments to use their data as they choose and are required to.
• Each city and jurisdiction is treated as unique.
• Preserving that uniqueness is part of the value proposition.
• Nick says the mission is to make every city more awesome in its own way.
• Serving the next customer requires serious integrity, moral compass, and core values.
• Nick says the next marginal customer must be cheaper to support by an order of magnitude.
• He says prior attempts by large organizations to do tailored public-safety software at affordable prices failed.
• Nick closes by saying the institution, not the vendor, has the power.