Weavel

@mwseibel thank you for the hunt! Hi Product Hunt makers 👋 I'm Andrew, co-founder of Weavel. I'm excited to introduce our prompt engineering automation platform for LLM apps @junpark_314 @hyunjiejung @tobykim . As foundation models get smarter, more and more tasks can be solved by prompting LLMs in the right way. But going from 80% to 100% is hard and requires days, even weeks of prompt engineering. We are building Weavel to help makers ship fast and with confidence, by automating the hassle of manual trial-and-error of prompt engineering with LLM-based algorithms. Weavel is designed for developers, data scientists, and AI enthusiasts who need to maximize efficiency and accuracy in their LLM apps. Weavel delivers the prompts with highest accuracy, 50x times faster than humans - optimized prompts surpasses DSPy-optimized prompts by 4% and CoT by 7%. We are rethinking prompt engineering to make the process fast, accurate, and reliable. Here's how we help the engineering cycle: 💾 Dataset curation: - Add one line of code to log LLM calls in your product. We automatically curate datasets from the logs - user representative dataset and out-of-distribution dataset for edge cases. - Enrich your dataset with LLM call logs from production. 🧪 Prompt optimization: - From your base prompt and provided dataset, Ape (our AI prompt engineer) will iterate on diverse prompts, and find the mixture of instructions and few shot examples with the highest scores. - It takes only 4 minutes, and improves accuracy by an average 20% 🚀 We'd love to hear your thoughts and feedback. Thanks for checking out Weavel! PS1: The code that powers Ape (AI prompt engineer) is open source at https://github.com/weavel-ai/Ape PS2: We are working on a prompt engineering playground that also automatically curates datasets from the runs - will launch soon! Real soon!!

Congrats on the launch, @aschung01 and the team! Automating prompt engineering can truly streamline the AI development process. I’m intrigued by the 20% accuracy boost—how does the platform ensure consistency across different datasets and models? Can't wait to see the playground feature in action!

Great! Congratulations! Many systems optimise assistant, RAG, and other frequent use situations, but few optimise AI Agent instructions and prompts.

This sounds really interesting! How does Weavel handle different languages or contexts for prompts? I'm curious if it's as effective across various domains or just limited to specific ones. Any insights on the user experience too?

Hey Andrew, How does the platform adapt to different LLM models, or is it focused on specific ones? Congrats on the launch!

This is a fantastic solution, @aschung01! Automating the tedious trial-and-error of prompt engineering could significantly speed up development. I’m especially interested in the dataset curation feature—how does it ensure the datasets are comprehensive and representative? Are there any plans to integrate more advanced features like real-time optimization or interactive tweaking?

Nice project @aschung01 and the team. Good luck with launch today!