LeadPilot AI is an autonomous AI sales employee designed to handle the messy, multi-step work that begins when a sales lead arrives. Instead of simply generating text, LeadPilot understands a customer's natural-language requirement, evaluates purchase intent, takes tool-based actions, maintains persistent memory, and determines the next sales action.
Features and Functionality: LeadPilot extracts customer requirements such as location, property type, budget, and timeline; scores and classifies leads as HOT, WARM, or COLD; searches matching property inventory; identifies suitable properties; stores lead and qualification information in Firestore; schedules follow-ups; and prepares personalized outreach. A Streamlit dashboard provides visibility into the complete workflow and runs on Google Cloud Run.
Technologies Used: The system uses Gemini 3.6 Flash for reasoning and natural-language understanding, Google Agent Development Kit (ADK) for agent orchestration, Python for implementation, Cloud Firestore for persistent business memory, Cloud Run for deployment, Streamlit for the dashboard, and Cloud Build/Artifact Registry for deployment infrastructure.
Other Data Sources Used: The prototype uses a demonstration property inventory containing attributes such as location, configuration, price, and possession timeline. Firestore provides persistent workflow data.
Findings and Learnings: The project demonstrated that the value of an AI agent comes from its ability to understand → reason → act → remember → follow up, rather than simply produce responses. We learned that reliable agentic applications require not only a capable model, but also tools, persistent state, cloud infrastructure, and clearly defined workflows. LeadPilot demonstrates how AI can reduce repetitive sales operations while keeping humans focused on higher-value customer interactions.
Built With
- cloudbuild
- cloudfirestore
- gemini
- googleadk
- googlecloudrun
- streamlit
Log in or sign up for Devpost to join the conversation.