impasse - master negotiation through practice

Inspiration

Platforms like LeetCode exist to prepare people for interviews, technical and behavioral alike Entrepreneurship much more common, thriving in startup culture requires strong negotiation skills Difficult to train negotiation skills because it requires another party Developed Impasse to solve this problem, with the goal of giving users a significant advantage in business deals, salary raises, and other high-stakes situations

What it does

Simulates a video call between you and an opposing party in a dealmaking situation (contract debates, car buying, fantasy football trade, etc.) Integrated AI coach that will give you helpful tips in order to improve negotiation skills in real time Post game analysis with things to improve your bargaining abilities through giving missed opportunities, revealing information, key lessons, and an overall score. Library that holds your previous recordings and analysis that came with it and an online database for all of the other recordings where you can post your own recording too

How we built it

Developed and engineered agentic prompts for Groq, resulting in opposition, coach, scenario generation, and post-mortem analysis agents Built a robust FastAPI backend with websocket connection capabilities, and included database routes for storing and retrieving individual negotiation sessions Designed and implemented a web interface in Next.js, and set up a workflow to generate scenarios, let the user take notes and take coaching hints, call with our opposition agent with live audio streaming and conversation, and generate detailed analysis on the session with insights for use in the future

Challenges we ran into

Connecting the agents and the frontend via web sockets on the backend and streaming audio back and forth was difficult since none of us had experience with it AI agents would break whenever reading online data for research and the text to speech wouldn’t interpret it well

Accomplishments that we're proud of

AI agent and user having constant communication through a mock video call Text to speech and speech to text working well with many varying talking styles and accents

What we learned

Practical application of agents in the web Speech to text and text to speech patterns with agentic purposes Combining different storage from the cloud to an SQL framework Using websockets to keep constant flow of data to cycle everywhere

What's next for impasse

Implementing emotion detection to better give advice for the user and more realistic communication by the debating AI Creating different AI agents for video calls that exposes the user to different types of negotiators so the user can be ready for real world scenarios