Dynamove

Overview

Dynamove is a voice-powered, AI-driven in-vehicle concierge copilot built for the Weavehacks 2025 hackathon. It enables users to search the web, make real reservations, and check their schedule using natural language—either by voice or text. The system leverages advanced LLMs, real-time web search, and browser automation to deliver a seamless, verifiable interactive experience, designed to enhance safety and convenience on the road.

Inspiration

The inspiration for Dynamove stemmed from the common frustrations experienced during long road trips and daily commutes. We envisioned a truly intelligent co-pilot that could handle complex, multi-step tasks – like booking a table at a specific type of restaurant, finding unique local attractions, or even checking personal appointments – all without the driver ever needing to touch a phone or navigate a cumbersome interface. The goal was to minimize distraction and maximize the joy of the journey, bringing the full power of the web safely into the vehicle.

What it does

Dynamove transforms the in-car experience by providing:

• Intelligent Restaurant Discovery & Booking: Users can ask for highly specific restaurant types (e.g., "authentic, family-run Mexican with mole and outdoor seating") and Dynamove, using real-time web search and automation, will find suitable options, display menus, and even book a reservation directly on the restaurant's website.

• Dynamic Point of Interest Discovery: When faced with unexpected detours or simply seeking adventure, Dynamove can proactively identify unique roadside attractions or scenic viewpoints along the updated route, providing details and seamlessly adding them as stops.

• Personalized Schedule Management: Users can quickly inquire about their upcoming appointments, such as a "dentist appointment," and Dynamove will access their Gmail to confirm details, providing critical information without visual distraction.

• Seamless Navigation Integration: All discoveries and bookings are seamlessly integrated with Google Maps, providing turn-by-turn directions to the chosen destinations.

• Natural Language Interaction: Powered by Gemini Live, Dynamove understands complex, conversational queries, making interactions feel intuitive and human-like.

How we built it

Dynamove is built on a robust architecture designed for responsiveness, intelligence, and real-world interaction:

• Google ADK (Android Development Kit): Serves as the foundational platform within the vehicle, providing a stable and secure environment for the application.

• Gemini Live: The core conversational AI and Large Language Model (LLM) that powers Dynamove's natural language understanding, intent recognition, and intelligent responses. It handles the user's voice input and crafts coherent, context-aware replies.

• MCP (Model Context Protocol) Servers: Our backend orchestrator. These servers manage the complex interactions between various tools like Gemini Live, Browserbase, and Exa. This architecture ensures that computationally intensive tasks and sensitive web automation occur securely and efficiently off the vehicle, minimizing in-car processing load and enhancing security.

• Browserbase: Integrated via our MCP servers, Browserbase is crucial for performing complex web automation tasks. This includes navigating websites, logging into accounts, filling out forms, extracting dynamic information (like restaurant menus or real-time hotel availability), and completing reservations directly on third-party websites that do not offer public APIs.

• Exa: Also integrated via our MCP servers, Exa provides powerful, real-time web search capabilities. It's used for deep, contextual information retrieval, such as finding specific restaurant reviews, identifying unique roadside attractions, or gathering detailed information about points of interest.

• Gmail API: Utilized to securely access the user's Gmail account (with explicit user consent) for retrieving information about scheduled appointments, such as dentist visits.

• Google Maps API: Provides the essential mapping and navigation services, allowing Dynamove to understand location context, display points of interest, and seamlessly update routes for bookings and detours.

Challenges we ran into

Building Dynamove presented several exciting challenges:

• Seamless API Integration: Orchestrating real-time communication and data flow between disparate services like Gemini Live, Browserbase, Exa, Gmail, and Google Maps, all while maintaining responsiveness for voice commands.

• Complex Browser Automation: Handling the dynamic and often unpredictable nature of web pages with Browserbase, including navigating various login flows, CAPTCHAs, pop-ups, and ensuring accurate data extraction from non-standardized web layouts.

• Maintaining Context: Ensuring Gemini Live could maintain conversational context across multiple turns and integrate information from various tools (e.g., remembering a preferred restaurant type while then checking its menu and booking availability).

  • Real-time Performance in a Vehicle: Optimizing the entire system for minimal latency to provide a fluid, non-distracting experience for the driver, especially given the off-car MCP server architecture.

  Security and Privacy: Implementing robust measures to protect user credentials for Browserbase and sensitive personal data accessed via Gmail, ensuring all interactions were secure and compliant.

  Designing for Minimal Distraction: Crafting a user interface that provides necessary information clearly and concisely without overwhelming the driver, relying heavily on voice output and intuitive visual cues.

Accomplishments that we're proud of

We are incredibly proud of several key accomplishments with Dynamove:

• Successful Multi-Step Web Automation: The ability to truly book a restaurant table or find detailed information on a specific website using Browserbase, going beyond simple API calls, is a significant breakthrough for in-car systems.

• Intelligent Spontaneous Discovery: Dynamove's capacity to find and integrate unique, context-aware points of interest (like specific Mexican restaurants or quirky roadside attractions) based on natural language queries, transforming unexpected detours into memorable experiences.

• Seamless Conversational Flow: The fluid and natural interaction with Gemini Live, which intelligently leverages the right tools (Exa for search, Browserbase for action, Gmail for schedule) without explicit user commands for tool selection.

• Robust Backend Orchestration: Building the MCP servers to efficiently manage and secure the complex interactions between all the powerful APIs, demonstrating a scalable and reliable architecture.

• Enhanced Driver Safety: By enabling complex tasks to be completed entirely hands-free and eyes-on-the-road, Dynamove significantly contributes to a safer driving environment.

What we learned

This hackathon pushed our understanding of integrating advanced AI with real-world web services:

• The Power of LLMs + Specialized Tools: We learned that while LLMs like Gemini Live are incredibly powerful for understanding intent, their true potential is unlocked when seamlessly integrated with specialized tools like Browserbase for action and Exa for precise information retrieval.

• The Necessity of Backend Orchestration: For complex, multi-tool interactions, a robust and secure backend (our MCP servers) is not just beneficial but essential for managing state, security, and performance.

• Browser Automation is a Game Changer: We gained deep insights into the capabilities and challenges of browser automation, realizing its immense potential for interacting with the vast majority of the web that lacks dedicated APIs.

• Designing for the "In-Car" Context: The unique constraints of a driving environment (e.g., limited visual interaction, varying noise levels for voice input) significantly influence design choices and require careful optimization.

What's next for Dynamove

The future for Dynamove is incredibly exciting! We envision expanding its capabilities to become an even more indispensable road trip and daily driving companion:

• Proactive Suggestions: Leverage learned user preferences and real-time data (weather, traffic, calendar) to proactively suggest detours, stops, or tasks before the user even thinks to ask.

• Pet-Friendly Travel Planning: Full integration for finding and booking pet-friendly accommodations, vets, and parks along a route.

• Vehicle Maintenance & Diagnostics: Deeper integration with vehicle diagnostics to proactively suggest service, find parts, or even schedule repairs via Browserbase.

• Personalized Shopping & Errands: Assisting with spontaneous shopping needs, finding specific items at nearby stores, and even initiating click-and-collect orders.

• Enhanced Emergency Assistance: More sophisticated voice-activated emergency protocols, including contacting roadside assistance or emergency services with pre-populated details.

• Integration with Smart Home: Allowing users to manage smart home devices (e.g., adjust thermostat, turn on lights) as they approach or leave their home.

Dynamove is just the beginning of a truly intelligent and seamlessly integrated in-vehicle experience.

Built With

• browserbase
• exa
• google-adk
• streamlit