Negotiation Copilot

Inspiration

Imagine having a trusted partner by your side, guiding you through every twist and turn of a negotiation. In today’s fast-paced B2B and eCommerce landscape, deals aren’t one-off events—they're evolving conversations. Our app goes beyond basic advice by anticipating counteroffers, exploring multiple strategies, and planning moves well ahead. This means you’re never caught off guard and always feel prepared for the unexpected.

What it Does

Negotiation Copilot is an AI-powered assistant that helps you navigate complex, multi-turn negotiations. It simulates realistic conversation scenarios and suggests the best next move to help you secure better deals—whether you’re discussing sales terms, contracts, or salary offers. Essentially, it gives you that extra edge by breaking down the negotiation into manageable steps and guiding you through each one.

How We Built It

We built Negotiation Copilot using a modern web stack that brings together several powerful tools:

Front End: We used React and Next.js to create a smooth, interactive user interface.
Back End: FastAPI runs our simulations and processes negotiation data efficiently.
Authentication: Stytch handles user login and session management seamlessly.
NLP Integration: Together.ai generates intelligent negotiation responses that mimic real-life interactions.
Negotiation Engine: Our core is a Monte Carlo Tree Search (MCTS) algorithm that simulates different conversation paths using a simple breadth-first search.
Performance Tracking: We integrated Weights & Biases to monitor MCTS expansions and help us optimize our scoring and decision-making processes.

Challenges

Bringing together so many different tools was no small feat. We had to ensure that data flowed smoothly between our front end and back end, and fine-tuning our MCTS logic while keeping computational demands in check was challenging. Adjusting the scoring prompts to reliably evaluate outcomes took many rounds of testing, and debugging across our entire stack taught us a lot about handling complexity under pressure.

Accomplishments

We’re proud of creating a prototype that not only meets hackathon requirements but also delivers tangible, real-world value. By blending advanced search algorithms with state-of-the-art LLM capabilities, we managed to build a tool that transforms negotiations into strategic, step-by-step conversations. The seamless integration of sponsor tools and our use of Weights & Biases for fine-tuning are achievements we believe set us apart.

What We Learned

Throughout the development process, we discovered that multi-turn simulations provide a far richer context than simple, one-off responses. Iterative testing and continuous refinement were crucial to improving our scoring mechanism. Using Weights & Biases, we gained deep insights into which negotiation paths were most promising, and this helped us to constantly tweak and improve our approach. Ultimately, this project has reinforced the importance of balancing technical innovation with real-world usability.

What's Next

Looking ahead, our plan is to add real-time negotiation tracking so the tool can adapt dynamically during live interactions. We’re also exploring how to expand its functionality to other scenarios like hiring and personal salary negotiations. Our long-term goal is to evolve Negotiation Copilot into a fully autonomous assistant that learns from every interaction, continually refining its strategies to deliver even more effective negotiation outcomes.