Calibrtr's guide to Chatbots

Guide hero.png

Chatbots are one of the easiest and most widespread applications of generative AI right now. These are simple, or not-so-simple, conversational interfaces which can be tailored for a whole host of usecases. Here at Calibrtr we’re really excited about the potential for developing tools with generative AI. But we also passionately believe that we need the right tools to really bring that potential to life. With that in mind, this is our guide to chatbot prompting, evaluation, and how to choose the right LLM to underpin your system.

Quick navigation:

Why would I want a chatbot?

Chatbots are the most accessible of the applications for generative AI at the moment. But that doesn’t mean that they have to be simple. Here’s a few ideas for chatbot use-cases:

  • Customer service bots
  • Website queries
  • Conversational data analysis
  • Corporate policy and knowledge banks
  • Virtual friends and partners
  • Booking agents
  • Sales and product customisation agents
  • Conversational programming and design tools
  • Tax, medical and legal advice

How do I build a good one?

There’s a lot of off-the-shelf chatbot services out there. But if you want to build your own, we have useful guides to some of the aspects of chatbot design:

What type of LLM do I need for my chatbot?

So, you want a chatbot. But what kind of LLM do you need: a LLM you buy as a service from a cloud AI, like ChatGPT, or one you run yourself, either on your own hardware or in the cloud? And how can you use techniques like RAG, Fine-Tuning and Prompt Engineering to make your products do what you want?

Type_of_LLM_hero.png

Writing great prompts for chatbots

The main way to direct a chatbot is through prompts. This guide goes through how to design prompts to make sure that your chatbot is doing what you want it to.

Writing Prompts Hero.png

Evaluating Chatbots.

Once you’ve got your first iteration of the perfect chatbot, how do you test its performance, both in terms of how it responds to the prompt and how it performs in the real world.

Evaluating chatbots hero.png

Key performance indicators for chatbots..

Want to get even more into chatbot evaluation? Here’s a starter for ten of Key Performance Indicators for chatbots to measure performance.

KPIs.png

Plus a bonus guide to reducing the costs of your LLM deployments.

Running costs hero1.png

Want to try out our services?

Calibrtr offers a Generative AI cost management and performance review platform, with tools to forecast and manage costs, build and evaluate prompts, experiment with different models and do A/B testing of prompts, monitor performance and build in human-in or out of-the-loop approvals and reviews. We’re currently in Beta- get in touch with us to find out more!

Our limited beta program is open

If you'd like to apply to join our beta program, please give us a contact email and a brief description of what you're hoping to achieve with calibrtr.

Please provide a valid email address
Thank you
Please let us know how calibrtr will help you
Thank you

Frequently Asked Questions

Chatbots can be used for a variety of applications, including customer service, website queries, conversational data analysis, corporate knowledge management, virtual companions, booking agents, sales and product customization, conversational programming tools, and even providing tax, medical, and legal advice.

When choosing an LLM for your chatbot, consider whether to use a pre-built service like ChatGPT or host an LLM on your own hardware. Evaluate factors such as the need for techniques like Retrieval Augmented Generation (RAG), fine-tuning, and prompt engineering to meet your specific requirements.

Effective chatbot prompts should clearly define the chatbot’s personality, provide positive and negative instructions, set guidelines for conversation openings and closings, and offer relevant context. Including examples and demonstrations in your prompts can also help guide the chatbot's responses.

When evaluating a chatbot, focus on whether it meets the intended design and performance criteria, including if it responds correctly to prompts, maintains the desired tone, and effectively performs in real-world scenarios such as improving user satisfaction and operational efficiency.

Key performance indicators for chatbots may include user satisfaction, response accuracy, efficiency in handling queries, lead generation, cost savings, and overall impact on brand experience. These KPIs help measure the effectiveness and value of the chatbot.

To reduce costs, consider using compression algorithms to shorten prompt sizes, perform A/B testing to optimize prompts, and use tools like Conversation Replay to refine interactions. Additionally, experiment with different models to find the best balance between performance and cost.

Calibrtr provides a platform for managing generative AI costs and performance. Our services include tools for cost forecasting and management, prompt building and evaluation, model experimentation, A/B testing, performance monitoring, and implementing human-in or out-of-the-loop reviews. We are currently in Beta, so get in touch to learn more.