Collaboration between developers and subject matter experts on prompts was broken
Creating differentiated LLM products often requires input from subject matter experts, in particular on AI prompt engineering.
At Twain, linguists own prompt engineering but the process was inefficient and error-prone: prompts were stored in code denying linguists direct access and forcing them to rely on engineers for any modifications. Whenever the linguists wanted to evaluate Twain’s LLM features, the engineering team had to fetch model outputs from its database and manually share them with the linguists.
The process required significant resources from the engineering team and considerably slowed down product development. Recognizing these inefficiencies, Twain sought a solution that would empower linguists to independently tweak prompts and assess their models, eliminating the need for continuous engineering intervention and extensive technical know-how.
Humanloop provides an interactive development environment for gen AI applications
Humanloop came as the answer to Twain's challenges. Through it, Twain overhauled its LLM development, detaching prompt engineering from core application development. Linguists, using Humanloop, now possess the autonomy to modify prompts and adjust model parameters, rolling out updates without requiring engineering input.
This revamped process has led to significant time and cost savings: Humanloop helps Twain save 6-8 engineering hours per week, 5-10 prompt engineering hours per week and 2 days of data science time for each new prompt in its AI product team. This translates to ca. $70,000 cost savings per team per year.
In addition Humanloop has proven to be a powerful tool for evaluating and monitoring Twain’s LLM applications, which again was previously done manually. As all data related to Twain’s LLM applications - prompts, generations, feedback and additional metadata - is stored on Humanloop, the team can easily evaluate its LLM applications and quickly identify opportunities for product improvement.
Humanloop saves Twain weeks of engineering work and makes complex AI features possible
With Humanloop as a strategic tool, Twain has refined its LLM application development and reduced the time and cost required.
This helped Twain to significantly expand its product offering: Twain's outreach capabilities now extend beyond email to LinkedIn and have incorporated various stages of communication, including follow-ups.
This swift product development, traditionally a months-long endeavour, was achieved in mere weeks.