Vector: Real-Time Logistics Autopilot

Vector

InspirationWe’ve all experienced the "Airport Panic": driving at 65 mph, hands on the wheel, only to hear a chime that your flight is delayed or canceled. You can't safely pull over, and you definitely can't navigate a booking app. My inspiration for Vector was to turn the AI from a passive "search engine" into an active "Logistics Autopilot." I wanted to build an agent that doesn't just deliver bad news but arrives at the conversation with the solution already in hand.

What it doesVector is a voice-first, low-latency agent powered by the Gemini Live API. While the user is driving and discussing a travel disruption, Vector is simultaneously:Scanning: Real-time flight data for the next three available departures.Calculating: New ETAs based on live traffic data via Google Maps.Executing: Preparing Uber bookings and hotel reservations.It handles complex, non-linear logic. If a user says, "Wait, check the other airport instead," Vector instantly pivots its API calls mid-sentence without losing the context of the original crisis.

How we built itVector is built on a Parallel Execution Architecture. We utilized the Gemini Live API's multimodal capabilities to process voice intent while triggering asynchronous tool-calls.Backend: A Node.js environment orchestrating multiple travel APIs.The "Brain": Gemini 1.5 Pro manages a high-density context window to keep track of "Plan A, B, and C" simultaneously.Logic: We implemented a "Confidence Scoring" system for tool-calling. If the agent finds a flight match $> 90\%$, it prepares the transaction for a simple verbal "Go" from the user.

Challenges we ran intoThe primary challenge was Latency Jitter and Context Interruption. In a real-world driving scenario, a user might interrupt the agent or provide fragmented information. We had to fine-tune our function-calling triggers to ensure that the agent started searching for hotel availability the moment it heard the word "canceled," rather than waiting for the user to finish their sentence. Managing state across these rapid-fire interruptions required a robust Redis-backed session manager.

Accomplishments that we're proud ofWe successfully achieved a "Zero-UI" experience. In our testing, users were able to rebook a flight, schedule a pickup, and notify their business contact of a delay—all within a 90-second voice conversation—without ever taking their eyes off the road or touching their phone. We also successfully integrated a secure "Voice-Match" authentication layer for processing payments.

What we learnedBuilding Vector taught us that the future of AI isn't about better "chatting"—it's about Agency. We learned that low-latency voice is a game-changer for accessibility; by removing the screen, we actually increased the utility of the AI. We also discovered that "Proactive Suggestion" (e.g., "I found a flight in 2 hours, should I book the Uber now?") creates a significantly higher level of user trust than a reactive response.

What's next for Vector: Real-Time Logistics AutopilotThe next step for Vector is Multi-Agent Collaboration. We want Vector to reach out to other agents—like your office’s AI scheduler—to automatically shift meetings based on the new travel itinerary. We are also looking into integrating Veo-powered visual summaries that can be sent to the user’s HUD (Head-Up Display) or smartwatch for a quick glance once they've reached a stop.

Built With

• express.js
• next.js
• node.js
• react
• typescript