Procurement Negotiator

• 

  Architecture

Inspiration

Procurement negotiations are often time-consuming, repetitive, and heavily reliant on manual analysis. I was inspired to streamline this process using intelligent agents that can autonomously handle quote evaluation, benchmarking, and negotiation. My goal was to reduce turnaround time, minimize costs, and empower procurement teams with data-driven decisions.

What it does

Procurement Negotiator is an AI-driven system built using the Google Cloud Agent Development Kit. It automates the entire procurement negotiation flow:

• Parses supplier PDF/Image quotes
• Benchmarks quotes using market data (if available)
• Develops strategic counter-offers.
• Sends responses and tracks supplier replies
• Updates the dashboard in real-time with negotiation statuses

It intelligently decides whether to accept a quote or counter based on predefined business rules and analysis insights.

How we built it

I used the Google Cloud Agent Development Kit to create a set of modular agents:

• DocAgent: Extracts and parses PDF quotes
• AnalysisAgent: Benchmarks quotes using internal datasets
• StrategyAgent: Crafts negotiation strategies and counter-offers
• CommsAgent: Handles email communication with suppliers.( Drafts email)
• Dashboard: Displays real-time status and outcomes

Agents communicate through message-passing and are deployed on Google Cloud infrastructure for scalability and reliability.

Challenges we ran into

• Designing effective coordination between agents without race conditions
• Ensuring high-accuracy parsing of diverse quote formats
• Creating realistic benchmarking models with limited initial data
• Handling edge cases like unresponsive suppliers or non-standard pricing terms

Accomplishments that we're proud of

• Successfully automated (with human in loop) an end-to-end procurement negotiation workflow
• Built a modular, extensible system with reusable agents
• Seamlessly integrated with other GCP services like Firestore, StorageAccounts etc (as a persistent store)

What we learned

• Agent-based design is powerful for orchestrating complex workflows
• Small, well-scoped agents are easier to test and maintain
• Clear message contracts between agents are key to system stability

What's next for Procurement Negotiator

• Integrate with live supplier api and biquery for real-time quote ingestion and selection.
• Enhance strategy formulation with machine learning
• Add support for multi-party and multi-round negotiations
• Open-source the project for community feedback and collaboration
• MCP with email providers

Built With

• cloudrun
• cloudstorage
• firestore
• gemini
• python