Inspiration
The inspiration for the project was providing prompt templates that would answer provided questions effectively. Through genetic engineering, we are able to evolve and select effective prompts that are chosen in the process similar to natural selection.
What it does
The product provides a framework for optimizing prompts for a language model through genetic algorithm iterations. It uses a combination of crossover, mutation, insertion, and deletion to explore and exploit the solution space, and it evaluates prompt fitness based on accuracy and language model perplexity.
How we built it:
Using Python, Pytorch, Pandas etc. and the Mistral-7B-v0.1 Open-Source LLM model, we were able to develop a prompt engineering process similar to natural selection.
Challenges we ran into
Challenges we ran into include:
- Integrating Hypercycle metrics
- Achieving a Double Diamond development structure
Accomplishments that we're proud of
- Developing a functional, Automated Prompt Engineering product.
- Identifying a use-case for the project (Developing effective prompt templates for users, based on their input)
- Achieving an effective use of the Double Diamond Structure (Everyone was grounded in a common solution and problem)
- Enduring the hackathon's complete duration (from identifying requirements to pitching our product in front of potentially thousands.)
What we learned
- Integration of HyperCycle technology
- Genetic Engineering methodology
- Extensive utilization of Open-Source Methodology]
What's next for Team Soon - Automated Prompt Engineering
- Tailoring our solution to industry use-cases
- Devoting prompt templates to optimize prompt engineering for the world.
Log in or sign up for Devpost to join the conversation.