Multi-Agent System for Subsea Threats (M.A.S.S.T)

Status: Operational  Uptime: Active across all platforms

MASST is a multi-agent system built to detect and respond to subsea anomalies in real-time—like pressure drops, signal loss, or unexpected patterns that may indicate emerging threats.

About this project

MASST was originally developed to monitor and protect subsea infrastructure. It detects anomalies such as sudden pressure drops that may signal anchor strikes, environmental shifts, or sabotage.

Since its development, MASST has evolved into a more flexible system—capable of adapting to other high-stakes environments where real time sensing, intelligent analysis, and coordinated response are critical.

MASST began as a natural extension of my earlier work exploring how multi-agent systems could interact with complex historical documents—particularly those housed at UC Berkeley’s Institute of Governmental Studies Library. That experience shaped my thinking around distributed reasoning: how different types of intelligence can be brought together to make sense of dense, multi-layered material. It also deepened my interest in platform resilience.

In turn, I designed MASST across multiple platforms to reflect a more realistic and defensible architecture. In the face of system failures or external threats, having agents distributed across platforms creates not just functional redundancy, but adaptive strength.

Agent Roles at a Glance

MASST is built around a network of five core agents—plus one observer—each with a distinct role, platform and way of thinking. Rather than depending on a single, centralized brain, MASST distributes intelligence across agents that collaborate, hand off, verify, and adapt as conditions shift. In addition, a human Responder remains outside the core loop, stepping in only when judgment or escalation is required.
Here’s how the agents operate:

Sense & Detect (Signal Agent)

From signal loss to pressure changes, the Signal Agent listens for disruptions in underwater environments. It standardizes input data, adds context like timestamps and location, and forwards potential anomalies for deeper review.

Interpret & Extract Meaning (Interpreter Agent)

This agent receives raw data and makes it legible. It filters out noise, maps signals to known patterns, and prepares a clear interpretation—making sure every decision down the line starts from solid ground.

Classify & Prioritize (Classifier Agent)

Not all threats are equal. The Classifier Agent determines what kind of issue the system is facing—assigning type, severity, and urgency. It helps the system know what to ignore, what to escalate, and how to act next.

Forecast & Simulate (Look-Ahead Agent)

This agent thinks ahead. It explores the consequences of inaction, simulates possible outcomes, and suggests next steps based on long-term system health. It’s not just reactive—it’s strategic.

Supervise & Resolve (Supervisor Agent)

The Supervisor keeps watch across the entire system. It audits how agents are working together, flags inconsistencies, resolves conflicts, and decides when to log, escalate, or close a case.

Confirm & Escalate (Human Responder)

This is the human-in-the-loop. After all agents complete their roles, the Responder reviews flagged cases and decides if real-world action is needed. The final judgement stays human.

Log & Observe (System Ledger)

This is the system’s memory. While other agents think, act, and decide, the Ledger quietly records. It tracks how agents respond, what outcomes they produce, and when events occur — creating a live, queryable archive of the system in motion.

Who is MASST for?

MASST is designed for people working at the intersection of infrastructure, intelligence, and resilience. Whether you’re in subsea monitoring, multi-agent systems research, critical infrastructure, or simply curious about how intelligent systems might respond to real-world threats, MASST is built to spark dialogue. It’s open by design—meant to be explored, extended, or questioned. If you’re interested in testing ideas, collaborating, or just thinking aloud together, I’d love to connect.
Send a Message