GlideLogistics Assist

GlideLogistics, powered by Google's ADK, redefines logistics customer care. It effortlessly handles queries, booking, tracking, and KYC via a smooth, conversational AI. Logistics Care, now a breeze.

Comment 2

The Spark: Our Inspiration

Coming from the heart of the logistics domain, we've experienced firsthand the common pain points and friction customers face when seeking support. Whether it's a simple query about packaging, the cumbersome process of booking, or the frustrating steps of KYC, we recognized a significant opportunity to redefine the customer support experience. Our inspiration was simple: to build an intelligent solution that eliminates these frustrations, making logistics support truly effortless for everyone.

What it does

GlideLogistics Assist is an agentic AI-powered customer support solution designed specifically for the logistics industry. Built using Google's Agent Development Kit (ADK), our agent provides comprehensive and seamless support by:

  • Answering packaging queries: Leveraging Retrieval Augmented Generation (RAG), it accesses and synthesizes information from a specialized knowledge base to provide accurate and immediate answers to common packaging questions.
  • Creating bookings: Through an intuitive conversational flow, it collects all necessary information from the user and then seamlessly calls a mock backend API to create new logistics bookings.
  • Tracking bookings: Users can provide a tracking number, and our agent instantly fetches and presents real-time tracking updates using mock data from the backend.
  • Completing KYC: It streamlines the Know Your Customer (KYC) process by accepting uploaded PAN card images, using OCR (Optical Character Recognition) to extract relevant data, and then leveraging LLM capabilities to process and verify the information.

Building Our Agent: A Phased Approach

Our development process was iterative and highly focused:

  1. Knowledge Acquisition: We began by immersing ourselves in Google's comprehensive documentation and leveraging the wealth of information available on YouTube tutorials to grasp the core concepts of the ADK and related services.
  2. Small-Scale POCs: To solidify our understanding and test feasibility, we developed numerous small Proof-of-Concepts (POCs) for individual functionalities, such as a basic RAG query or a simple API call.
  3. From POC to Production: We meticulously converted these isolated POCs into integral parts of our agent, integrating them one at a time. This modular approach allowed us to build complexity gradually while ensuring stability.
  4. Continuous Refinement: Throughout the development cycle, we continuously refined the agent's logic, conversational flow, and integration points, striving for a seamless and highly effective user experience.

Navigating the Hurdles: Challenges Faced

Developing GlideLogistics Assist within a hackathon timeframe presented its own set of significant challenges:

  • Steep Learning Curve in Crunch Time: Mastering entirely new frameworks and APIs (ADK, Google LLMs, Vertex AI, Google Cloud) under a tight deadline was intensely demanding.
  • Hectic Work Schedules: Juggling project development with existing work commitments added pressure and required efficient time management.
  • Remote Team Collaboration: Coordinating efforts and maintaining seamless communication across a geographically dispersed team required disciplined planning and effective virtual collaboration tools.

Despite these hurdles, our determination to innovate and deliver a practical solution for logistics customer support propelled us forward, resulting in GlideLogistics Assist.

Technical issues

** Learning time - Given that we were all new to multi agent and ADK with A2A and MCP, there was quite some learning to be done ** Consuming the Agents on a conversational interface. We started with a simple SPA mode but couldn't move to conversational interface. ** Could deployment is pending. We managed to deploy one agent but could not integrate it with overall orchestrator within the deadline ** Other items which were on agenda but we couldn't complete - Session Management and Memory Management with external services, user authentication and personalization based on the same.

Accomplishments that we're proud of

Despite the challenges, we're incredibly proud of what we've achieved with GlideLogistics Assist:

  • Functional Multi-Capability Agent: We successfully built and integrated diverse functionalities (RAG, API calls for booking/tracking, OCR for KYC) into a single, cohesive AI agent.
  • Rapid Adoption of New Tech: Our team quickly adapted to and implemented Google's ADK and associated cloud services from a standing start, demonstrating strong learning agility.
  • Seamless User Experience (Mock): We've engineered a conversational interface that, even with mock backends, demonstrates how effortlessly a user could interact with logistics support.
  • Real-world Problem Solving: We've tackled common friction points in the logistics customer journey, proving the tangible value of agentic AI in this domain.

What we learned / Our Learning Journey

This project was a deep dive into new technologies for our team. While we had some prior exposure to Langchain and OpenAI, the Google ecosystem for agent development was a fresh frontier. We embarked on a steep learning curve, starting from the ground up to understand Google's Agent Development Kit (ADK), various Google LLMs, Vertex AI APIs, and relevant Google Cloud APIs. Almost every component, from foundational concepts to intricate implementations, was a new discovery that fueled our project's development.

What's next for GlideLogistics Assist

The hackathon is just the beginning for GlideLogistics Assist! Here's what we envision for its future:

  • Integration with Real APIs: Our immediate next step is to replace the mock APIs with live integrations to actual logistics booking and tracking systems.
  • Enhanced RAG Knowledge Base: We plan to expand the RAG knowledge base significantly, incorporating more detailed and nuanced packaging guidelines, regulations, and common FAQs.
  • Advanced KYC Verification: We aim to implement more robust KYC verification steps, including liveness detection and integration with government databases (where permissible and secure).
  • Personalization: We want to authenticate the user and then personalize the experience e.g. Users preferences from previous conversations, previous orders
  • Proactive : Developing features for proactive customer updates e.g., upcoming deliveries, proactive calling out potential delays, gentle reminders of invoice, etc.
  • Multi-modal Input/Output: Exploring support for voice input and output, as well as the ability to process more diverse document types for KYC and other processes.
  • Scalability & Deployment: Optimizing the agent for scalable deployment on Google Cloud, ensuring it can handle high volumes of customer interactions.

We believe GlideLogistics Assist has the potential to significantly enhance customer satisfaction and operational efficiency across the logistics industry.

Built With

Share this project:

Updates