Procurement Assistant

Initial Screen

Inspiration

Industrial procurement is a tedious and time-consuming process. Traditional manual approaches often involve significant paperwork, communication delays, and lack of transparency, making it difficult for companies to efficiently compare options and negotiate the best deals. We wanted to create a solution that leverages AI to streamline and automate this process, especially for high-value items like laptops, so organizations can optimize their purchasing decisions and reduce operational overhead. The final decision is always made by the human user—the AI only negotiates and analyzes on their behalf.

What it does

Procurement Assistant is an AI-powered platform that simplifies and accelerates procurement. The user specifies their procurement requirements—such as the need for laptops with specific configurations. The system then identifies the best-matching laptops from a synthetic vendor dataset, finds suitable vendors, and enables the user to select two vendors for negotiation. The negotiation agent then initiates parallel negotiations with both vendors, following best practices, to achieve the optimal deal. The process is transparent, interactive, and culminates with a comprehensive deal summary to help the user make an informed decision.

How we built it

We built Procurement Assistant using Python 3.12+, FastAPI for backend APIs, and Streamlit for the interactive frontend. The core intelligence is powered by Google ADK (Agent Development Kit), which enables us to create modular agents for matching, vendor selection, negotiation, and deal summarization. Synthetic datasets simulate real-world vendor offerings. The negotiation process is structured, with buyer and seller agents simulating realistic negotiation dynamics, and the results are presented in an easy-to-understand format for the user.

Challenges we ran into

Designing realistic negotiation logic between buyer and seller agents.
Ensuring smooth parallel execution of negotiations for multiple vendors.
Handling streaming responses and deduplication for real-time feedback.
Simulating realistic vendor data to mimic real-world procurement scenarios.
Managing CORS and API connectivity between the ADK server, CORS proxy, and frontend.

Accomplishments that we're proud of

Successfully automated the multi-vendor procurement negotiation process.
Created a modular, agent-based architecture that can be extended to other procurement categories.
Implemented real-time, streaming negotiation feedback with deduplication.
Developed a professional, user-friendly interface for both standard and streaming negotiation modes.
Simulated realistic negotiation outcomes and deal summaries for effective decision-making.

What we learned

The power and flexibility of modular agent-based systems for complex business workflows.
Practical aspects of AI-driven negotiation and procurement processes.
The importance of user experience in AI-powered business tools.
Technical challenges involved in integrating multiple backend services and handling real-time data streams.

What's next for Procurement Assistant

Integrating with real vendor APIs for live procurement scenarios.
Expanding the platform to support procurement across multiple product categories.
Enhancing analytics dashboards for deeper procurement insights.
Automating re-negotiation based on dynamic requirements or market changes.
Adding features to export negotiation summaries and reports in various formats (PDF, Excel).
Further improving agent intelligence for even better negotiation outcomes.