Explore the capabilities and limitations of current LLM-based software development agents by implementing one yourself.

For this hackathon, an AI agent is defined as a piece of software that autonomously solves a given task by repeatedly calling an LLM. The agent receives an unsolved task from the backend, writes code, executes the code, and submits a solution.

We will be participating in the Advent of Code, a yearly contest featuring 25 (times 2) coding puzzles to solve in any programming language. The goal of our hackathon is to solve as many of the 2023 puzzles as possible.

Participants are not allowed to directly write code to solve the puzzles. Instead, we must use agentic AI tools to achieve this.

• Only prompts, glue code, API integration, and agent logic (e.g., passing compile errors back to the agent) can be written.
• Utilize APIs from providers like OpenAI, Groq, Anthropic, Google, etc.
• Prompts must not be puzzle-specific.

The AI agent must be re-run from puzzle 1 after each modification without retaining memory of previous runs. It starts from scratch every time.

• Each puzzle is usually self-contained and not overly complex.
• The solution should fit within the context length of a conventional model and can be easily verified.
• The puzzles are given in text form, requiring no multi-modality.

A version of the Rust backend is deployed at http://32k.eu:8000/api. Availability is not guaranteed. See the "Bruno" REST client templates under bruno.

You will need to provide a Bearer Token with each request to authenticate (ask Beni).

For detailed information on how to build, run, and use the backend, see backend-rs.

A simple agent implementation can be found under simple-bot-py.