User's pain point
Developing LLM-based AI features is like a black box,
making evaluation and confident launches difficult.
Snowbell AI assists users in streamlining their LLM workflows, encompassing data preparation, prompt engineering, and evaluation & testing. With a comprehensive AI development toolkit, it makes building AI features more efficient and transparent.
At the beginning of 2023, my co-founder (an AI engineer) and I recognized a growing demand for a new approach to developing LLM-based applications. After conducting dozens of user interviews to gather and assess insights, we decided to build a platform that enables businesses to create LLM apps more efficiently and launch them with greater confidence.
As a co-founder, my responsibilities included:
Why we focused on this area ...
While my co-founder and I were developing AI software to enhance the drug development cycle (our previous startup idea ), we discovered a significant demand for LLM engineering tools, yet the market lacked options beyond basic data labeling tools. After three months of interviews and surveys to validate our assumptions, we decided to create a prompt engineering tool specifically designed to support AI feature development.
Snowbell AI MVP Roadmap
As co-founder, I oversaw the entire product and business development process—from validating opportunities and designing the MVP to launching, driving sales, and making post-launch improvements. This hands-on experience in validating product-market fit and sales gave me a deep understanding of the importance of designing for real user needs and how user insights drive product strategy.
User insights shaping our product strategy
Developing LLM-based AI features is like a black box,
making evaluation and confident launches difficult.
Empowering customers to confidently launch high-quality AI features.
Put users first,
keep it simple and intuitive,
be transparent and clear,
prioritize users with AI as support.
Define new features based on user's feedback
We gather feedback from user interviews and surveys throughout the product development cycle, prioritizing features with the highest business impact, alignment with user needs, and low engineering effort. This approach allows us to iterate quickly based on genuine user needs.
Key product features quick walkthrough.
Home
Users can quickly resume where they left off on the home screen, while the Activities panel on the right keeps them updated on relevant activities.
Projects
Side by side experiment list - List view
Each project includes an experiment list where users can browse metrics data for each experiment.
Side by side experiment list - Visualization view
The data visualization view lets users compare all experiments in seconds, helping them confidently choose the best-performing model to deploy.
Create a new experiment
User can conduct side-by-side comparisons of multiple prompts, parameters, models, and even model providers across a range of test cases.
Experiment detail page
Users can dive deeper into each experiment by reviewing individual test cases by query and accessing labeling as needed.
Prompt playground–Prompt editor
With the 'Add to Query Set' feature in the prompt editor, users can easily save valuable prompt variables they've created during editing, allowing for reuse in future queries.
Prompt testing–Prompt test
We found that traditional unit testing tools are lacking in prompt engineering. To address this, we developed an inline testing feature, enabling users to quickly test prompts on any query in the playground. This approach is both fast and cost-effective.
Query set editor
Invite people to projects
