• 

  Starting screen

• 

  During the conversation

• 

  Summary

• 

  askLio Simulation Enviroment

Sekundant

Sekundant is a real-time negotiation co-pilot that helps buyers make better decisions during high-stakes vendor conversations.

Inspiration

Negotiations are stressful because the most important moments happen fast. When a vendor says, “This is my final offer,” you have seconds to decide whether it is true, a bluff, or just a pressure tactic. We wanted to build an assistant that gives users expert-level negotiation guidance while the conversation is still happening.

What It Does

Sekundant analyzes buyer-vendor conversations in real time and gives short, actionable hints during the negotiation. It can detect suspicious claims, highlight inconsistencies, suggest tactical next moves, and summarize the conversation afterward.

Our prototype includes a live negotiation simulation, an AskLio-powered vendor, AI-generated hints, conversation storage, document upload support, and Claude-powered negotiation summaries.

How We Built It

We built Sekundant as a multi-service prototype. The negotiation UI communicates with a WebSocket simulation server, which sends messages to our AI connector. The connector stores conversation history, runs several specialized Claude prompts in parallel, and returns a final color-coded hint to the user.

A FastAPI backend manages companies, conversations, documents, and summaries. The broader product concept also connects to a native macOS app that captures and transcribes meeting audio, turning live calls into negotiation intelligence.

Challenges

The biggest challenge was making the AI useful in real time. A good negotiation assistant cannot produce long generic advice; it has to be short, grounded, and immediately usable. We solved this by splitting the reasoning into multiple AI agents for suspicion detection, tactical suggestions, and fact checking, then merging them into one concise recommendation.

What We’re Proud Of

We built an end-to-end prototype that can simulate a negotiation, analyze both sides of the conversation, and return live strategic advice. We are especially proud that the assistant does not just detect problems: it recommends what the user should do next.

What’s Next

Next, we want to improve document-based context retrieval, polish the macOS call experience, support more negotiation scenarios, and evaluate the quality of recommendations across real procurement, sales, and salary negotiation examples.

Built With

• bedrock
• fastapi
• mysql
• swiftui
• websockets