Port-Power Nexus #25

Inspiration

Our project was inspired by a looming environmental and logistical crisis at the Port of Long Beach. By 2035, California’s Clean Trucks Program mandates the replacement of approximately 22,000 diesel trucks with electric rigs. This transition presents a massive challenge: our current "dumb" grid cannot handle the surge in demand if thousands of trucks plug in simultaneously, potentially causing brownouts across Orange County. We wanted to build a solution for West Long Beach—one of the most diesel-polluted communities in the US—that moves beyond slow, centralized dispatchers.

What it does

Port-Power Nexus is a decentralized micro-economy designed to manage electric vehicle (EV) charging through an autonomous AI swarm. Instead of a central server, we envisioned a "high-speed stock market for electricity" where:

The Grid acts as the auctioneer, broadcasting dropping prices based on real-time California grid stress and renewable energy percentages.
Trucks act as autonomous day-traders, using LLM-driven reasoning to negotiate the best time to charge based on their specific battery levels and current market conditions.

How we built it

We developed a multi-layered architecture to bridge high-level reasoning with decentralized infrastructure:

ASI:One Chat Interface: A natural language entry point where a fleet manager can trigger the entire swarm with a single sentence.
Backend & Agents: Python-based agents (Orchestrator, Grid, Terminal, and Truck agents) communicating via the Fetch.ai uAgents protocol.
Intelligence: OpenAI API to provide LLM reasoning for each truck's bidding strategy.
Data & State: A Supabase database to maintain real-time state for auctions, truck status, and bay availability.
Frontend: A React dashboard built with Vite and Tailwind CSS, using Supabase Realtime to animate the live auction countdown and pulsing charging nodes.

Challenges we ran into

We utilized a multi-agent architecture powered by uAgents and Agentverse, which was our first time working with the Fetch.ai ecosystem. Navigating the learning curve of asynchronous agent communication was a significant hurdle. Furthermore, we had to pivot mid-hackathon when our primary ISO API source for tracking electricity price per kWh went down. We refactored our Grid agent on the fly to ingest data from an alternative source to ensure the auction logic remained functional.

Accomplishments that we're proud of

Autonomous Negotiation Engine: We successfully implemented a system where truck agents don't just follow static rules but use LLM reasoning to weigh factors like battery health and delivery deadlines against market prices.
Real-Time Decentralization: We built a functional "high-speed stock market" for energy that operates without a single point of failure using the Fetch.ai protocol.
Full-Stack Synergy: Synchronizing a Python agent swarm with a Supabase Realtime backend and a React frontend to visualize a complex, invisible process into a clean "Command Center."

What we learned

Agentic Orchestration: We gained deep technical experience with the Fetch.ai ecosystem and managing asynchronous communication between independent agents.
The Math of Micro-Economies: We dived deep into the mechanics of Dutch Auctions. To prevent grid surges, we modeled price decay functions where the price $P$ at time $t$ follows an exponential drop: $P(t) = P_0 \cdot e^{-kt}$, where $k$ is the decay constant representing grid urgency.
Pivot Resilience: We learned the importance of modular data ingestion. Swapping our price data source during an API outage proved that a decentralized system must also be a flexible one.

What's next for Port-Power Nexus

V2G (Vehicle-to-Grid) Integration: Transforming trucks into "mobile batteries" that can sell power back to the grid during peak stress.
Predictive Congestion Modeling: Integrating historical traffic data from the 710 freeway to help agents predict "Arrival at Port" times accurately.
Hardware Prototyping: Moving beyond the digital dashboard to integrate with actual EVSE (Electric Vehicle Supply Equipment) controllers to physically throttle or initiate charging sessions based on won bids.