Case study
Botler AI Assistant
Our in-house lab project: the AI assistant on the Flow State Works site, where we prove prompt-engineering and retrieval patterns before they ship in client work
Project overview
Botler is the conversational AI assistant embedded on the Flow State Works site — our in-house lab project. We use it to test prompt-engineering, context-management, and retrieval patterns in a low-stakes setting, so the assistants we build for clients launch on approaches we've already proven in production.
The challenge
Problem: Trying out new prompt-engineering, context-management, and retrieval techniques directly on client projects is risky — we needed a low-stakes, production-like environment to prove each approach before it ships in client work.
Technical implementation
Prompt engineering
Carefully crafted system prompts ensure consistent, helpful responses while maintaining the AI's personality and avoiding common pitfalls.
Efficiency optimization
We implemented token management and response caching to minimize API costs while maintaining fast response times.
Context management
Short term context history to prevent token limit issues in longer conversations.
Fair working conditions
Even though Botler is a few lines of code, they enjoy fair working conditions. Botler "retires to sleep" after 5 responses.
Key features
- Natural conversation flow with memory
- Optimized for marketing and analytics queries
- Real-time streaming responses
- Mobile-responsive chat interface
Need a system like this?
Tell us where work gets stuck — in one call you'll hear what it takes to remove it.
Technologies used
Business value
- Client assistants launch on patterns already proven in production
- A low-stakes lab for testing prompt, context, and retrieval approaches
- Token management and response caching keep API costs down