Empowering Non-Technical Teams with Prompt Engineering and PromptLayer at Speak

How Speak Empowers Non-Technical Teams with Prompt Engineering and PromptLayer

Jonathan Pedoeem May 16, 2024 3 min read

How Speak Uses Prompt Engineering and PromptLayer to Empower Product Teams

The following is a case study of how Speak uses PromptLayer .

Speak.com, a language-learning app backed by OpenAI’s Startup Fund, has seen massive growth, expanding from 1 to 11 markets within a year. As they scaled, Speak needed to find ways to enable their non-technical teams, particularly in content and product, to iterate on AI features without requiring constant engineering support. PromptLayer provided the solution, allowing Speak to systematize and operationalize content work to rapidly improve their AI. The impact was so significant that Speak’s entire content team was trained on using PromptLayer.

Curricular Content Production

Seung Jae, the Product Lead at Speak, led an effort to drastically scale the creation of fill-in-the-blank multiple choice questions, a core part of Speak’s language learning curriculum.

Challenge:

Generating large volumes of new curriculum questions, supporting numerous languages, and tailoring content to different learner skill levels.

Solution:

SJ used PromptLayer’s batch pipeline builder to create a 6-step prompt chain that took curriculum definitions as input and output formatted JSON that could be directly plugged into their app. The chain included steps for generating lesson groups, writing level-appropriate sentences, and converting to JSON.

First, SJ used PromptLayer to evaluate each step modularly. Once satisfied, he triggered a full batch run in the morning and went out to get a coffee. In an hour the results were ready to be plugged into the product.

Results:

“Using PromptLayer, I completed many months’ worth of work in a single week. It empowered me to drastically scale our content creation process, going from curriculum outlines to app-ready content that users could engage with immediately.”

Seung Jae Cha, Product Lead

Product Development

Speak’s “Magic Wand” feature allows users to easily generate role-based scenarios for freeform English conversation practice.

Challenge:

Developing the right prompts to power this feature required extensive iteration and testing. However, the engineering team ill-equipped and too occupied to focus on perfecting the AI voice.

Solution:

Using PromptLayer, Speak’s PMs were able to rapidly test prompt variations, iterate, compare results, and run evaluations, all without needing to involve engineering.

Results:

International Expansion & Business Operations

As Speak expanded from 1 to 11 markets in a year, localization of customer support responses was a key focus.

Challenge:

Generating high-quality localized customer support responses for each new market.

Solution:

Speak’s business operations team used PromptLayer to create prompts that, combined with historical support data, could generate responses in the new languages that were then refined by localization teams. These contractors would start refining translations at 80% complete instead of writing them from scratch.

Results:

Key Takeaways

By enabling Speak’s content, product, and bizops teams to independently implement and iterate on prompt engineering workflows, PromptLayer delivered:

PromptLayer put the power of prompt engineering directly into the hands of the teams closest to Speak’s content and customers, resulting in a more efficient, more agile, and more scalable AI-driven business.