LLM-Trader

Inspiration

What would be the optimal strategy if we were to directly negotiate and trade with an LLM? This question sparked the idea.

What it does

Everyone competes daily to see how expensively they can sell a newly updated item to an LLM through a prompt. The LLM has a keen eye and can somewhat predict the price of the item. Any storytelling is welcome, as the item is an unpredictable object that has drifted through countless spacetimes and timelines. Try to dispose of it at the highest possible price!

How we built it

Pixel-image-based items and descriptions, newly generated daily via an Ollama + Stable Diffusion server, are uploaded to Supabase. On the webpage, an LLM Agent, powered by Gemini, acts as a 'judge,' receiving prompts and calculating an appropriate price.

Challenges we ran into

Integrating various models into a single webpage proved complex.

Accomplishments that we're proud of

However, we were able to quickly build the framework using Bolt and then refine it.

What we learned

For small-scale projects, it can be better to build first and then refine.

What's next for LLM-Trader

Beyond just selling items, we are also considering a system where users can buy items cheaply from the LLM Trader's store and trade among themselves.

Built With

• bolt
• cursor
• gemini
• ollama
• stablediffusion
• supabase