[NEW] Structured Prompt Outputs & Variable Pathing

Voiceflow · Intermediate ·✍️ Prompt Engineering ·1y ago

Key Takeaways

Voiceflow's new features, Structured Prompt Outputs and Variable Pathing, enable users to build AI agents with more efficient variable management and specific JSON output formats from Open AI LLMs, such as GPT 4.0.

Full Transcript

voice loow just got more powerful we've released two features that let you build AI agents without needing to create a gazillion variables and ones that make sure open AI llms return data in a specific Json format rather than me telling you about these features let me show you first up structured outputs structured outputs let you force an llm to return data in a specific format for example if you're building a sentiment classifier you could force your AI agent to return the emotion V Texas expressing and how sure it is of that fact so let's try using them now I'm going to add a set step to my canvas logic and a set and I have already built myself a nice little prompt so I'm going to set this using a prompt and the prompt is going to be get City info so let me show you some info about this prompt I'm literally just saying you're an AI agent get City from your knowledge and I'm passing in a city at no point in my prompt am I actually telling it what output I'm expecting but if I open settings you can see I'm using GPT 40 structured outputs are supported in GPT 40 and full om many at release and we have a new Option here Json output that's enabled if I give it a click you can see here that we have defined an output format now this output format follows a subset of open AI syntax and it supports these types but heads up it doesn't support nesting you can learn more at this link so we have defined an output that that has two properties we're expecting a city name which is a string and a population which is a number so now we set this output format we can then use that in other steps so let me close out of here and I'm going to make sure that I'm setting a variable so City info to the output of this prompt awesome so now we have an output stored in a variable but until today you need to create multiple variables one for each property that's being outputed you case city name and population but no more thanks to a feature called variable pathing Nina built it and she's going to tell you a bit about it well actually shout out to Pedro who build this feature as well so now that you have the city info stored in a single variable you can base your further Logic on its properties I'm going to add a condition of Tye business logic here and I want to check if the population of the city returned in City info is bigger than 5 million so I'm hitting tab after City do population enter is greater than let's say 5 million 1 2 three six okay if it is I'm going to say whoa city info. city is a big city or otherwise seems like a small City to me okay now that we've built it let's try it out together I'm going to hit start and it prompts me for a city and I would much rather be in Tokyo right now so I'm going to say Tokyo and as you can see it says it's a big city and indeed the population of Tokyo is more than 5 million that was possible thanks to the structured output returning all the city info to this city info variable and US reading the de population property of it and comparing it to the number that we set now I'm going to try to reset this test and try Toronto and indeed it seems like a small city which we already knew pretty cool right oh and Jac has one more thing to show you thanks CA I want to show you that structured outputs support more than just prompts for example I have this API that returns a bunch of posts from my website all the posts are in laddin because uh reasons instead of capturing each piece of data about a post manually it's title ID and body I can just store the entire object about a post in a variable using variable pathing then my AI agent can tell me all about the latest post on my website can even do a little bit of sentiment analysis to determine how I was feeling in that in this case it was a completely neutral story with variable paing you no longer need to create multiple variables to store properties about the same piece of information for example a user and with structured outputs you can be sure that your calls to llms are returning the data that you would expect good Lu the building and let us know in the comments down below what you plan to use these new beaches for

Original Description

Discover new ways to build with structured outputs and variable pathing. Read the changelog here: The fastest way to build, manage, and deploy AI agents. Use Voiceflow to design, test, and launch chat or voice AI agents — together, faster, at scale. Join our Discord community 👾 https://link.voiceflow.com/community Kickstart your next project with our templates 🚀 https://link.voiceflow.com/marketplace-youtube Our Links 🔗 👉 Start building today: https://www.voiceflow.com/?utm_source=youtube&utm_medium=organic 👉 Docs: https://docs.voiceflow.com/ 👉 Subscribe: https://bit.ly/3am22nf 👉 Twitter: https://bit.ly/2xrXZqV 👉 LinkedIn: https://www.linkedin.com/company/voiceflowhq/ 👉 Publication: https://www.voiceflow.com/blog?utm_source=youtube&utm_medium=organic
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Playlist

Uploads from Voiceflow · Voiceflow · 36 of 60

1 The Context Window Paradox with LLMs
The Context Window Paradox with LLMs
Voiceflow
2 Intercom to Voiceflow: why Nick, Head of CX @ Roam made the move #ai #customersupport #chatbot
Intercom to Voiceflow: why Nick, Head of CX @ Roam made the move #ai #customersupport #chatbot
Voiceflow
3 Biggest Challenge with Intercom - why Nick moved to Voiceflow #ai #chatbot #automation #intercom
Biggest Challenge with Intercom - why Nick moved to Voiceflow #ai #chatbot #automation #intercom
Voiceflow
4 Save 30 HOURS a week automating your customer support #ai #customersupport #automation #voiceflow
Save 30 HOURS a week automating your customer support #ai #customersupport #automation #voiceflow
Voiceflow
5 NLUs vs. LLMs - are NLUs dead? #nlu #llm #gpt #ailearning #agent #voiceflow #largelanguagemodels
NLUs vs. LLMs - are NLUs dead? #nlu #llm #gpt #ailearning #agent #voiceflow #largelanguagemodels
Voiceflow
6 Build a GPT4 Vision AI Assistant #business #gptv #gpt4 #ailearning #ai  #developer
Build a GPT4 Vision AI Assistant #business #gptv #gpt4 #ailearning #ai #developer
Voiceflow
7 How does an AI model search through information?
How does an AI model search through information?
Voiceflow
8 Gamechanging Zendesk app that summarizes your tickets #customersupport #ai #zendesk #voiceflow
Gamechanging Zendesk app that summarizes your tickets #customersupport #ai #zendesk #voiceflow
Voiceflow
9 Three reasons why your business shouldn't build a custom LLM
Three reasons why your business shouldn't build a custom LLM
Voiceflow
10 Do we still need Conversation Designers?
Do we still need Conversation Designers?
Voiceflow
11 LLMs have changed Conversation Design forever...  #ai #generativeai
LLMs have changed Conversation Design forever... #ai #generativeai
Voiceflow
12 Conversation Designer or Agent Designer? The Future of AI Automation Design #ai #generativeai
Conversation Designer or Agent Designer? The Future of AI Automation Design #ai #generativeai
Voiceflow
13 What's New in Voiceflow | March Feature Releases
What's New in Voiceflow | March Feature Releases
Voiceflow
14 Voiceflow AI Agency Panel: Start an AI Agency that's Built to Last
Voiceflow AI Agency Panel: Start an AI Agency that's Built to Last
Voiceflow
15 9 Tips for Starting and Scaling Your AI Agency
9 Tips for Starting and Scaling Your AI Agency
Voiceflow
16 What's New in Voiceflow | April Feature Releases
What's New in Voiceflow | April Feature Releases
Voiceflow
17 How to Scale Your AI Agent | Crawl, Walk, Run
How to Scale Your AI Agent | Crawl, Walk, Run
Voiceflow
18 The most important thing Large Language Models can do
The most important thing Large Language Models can do
Voiceflow
19 Three ways to use LLMs in your company
Three ways to use LLMs in your company
Voiceflow
20 5 Conversational AI Frameworks for AI Agents
5 Conversational AI Frameworks for AI Agents
Voiceflow
21 Voiceflow is a Customizable AI Platform
Voiceflow is a Customizable AI Platform
Voiceflow
22 Know your AI Agency Customers
Know your AI Agency Customers
Voiceflow
23 The Future of AI is Custom Interfaces
The Future of AI is Custom Interfaces
Voiceflow
24 The Overnight AI Agency Gambit
The Overnight AI Agency Gambit
Voiceflow
25 Introducing Tabular Data Support | June Feature Releases
Introducing Tabular Data Support | June Feature Releases
Voiceflow
26 Getting Started with Voiceflow APIs
Getting Started with Voiceflow APIs
Voiceflow
27 An AI Coach that Drives Leads and Financial Literacy
An AI Coach that Drives Leads and Financial Literacy
Voiceflow
28 Unlocking LLM Accuracy — Let It Cook!
Unlocking LLM Accuracy — Let It Cook!
Voiceflow
29 Speed Up Your AI Agent — Make Concurrent API Calls!
Speed Up Your AI Agent — Make Concurrent API Calls!
Voiceflow
30 Save Big with Automation — Cutting Costs Effectively
Save Big with Automation — Cutting Costs Effectively
Voiceflow
31 Multimodal Projects, LLM Entity Extraction, Cheaper Tokens, and More!
Multimodal Projects, LLM Entity Extraction, Cheaper Tokens, and More!
Voiceflow
32 Add a phone number to your AI agent on Voiceflow
Add a phone number to your AI agent on Voiceflow
Voiceflow
33 Top 5 Voice AI Agent Best Practices
Top 5 Voice AI Agent Best Practices
Voiceflow
34 Voiceflow 2024 Recap
Voiceflow 2024 Recap
Voiceflow
35 Build Voice AI Agents with no-code in Voiceflow
Build Voice AI Agents with no-code in Voiceflow
Voiceflow
[NEW] Structured Prompt Outputs & Variable Pathing
[NEW] Structured Prompt Outputs & Variable Pathing
Voiceflow
37 This AI agency's Project for a Local City Hall Drives over 11,000 Monthly Interactions #aiagency
This AI agency's Project for a Local City Hall Drives over 11,000 Monthly Interactions #aiagency
Voiceflow
38 Your AI Interface is More Important than the Content | Humans Talking Agents Episode 1
Your AI Interface is More Important than the Content | Humans Talking Agents Episode 1
Voiceflow
39 The Future of AI Automation Agencies | Humans Talking Agents Episode 2
The Future of AI Automation Agencies | Humans Talking Agents Episode 2
Voiceflow
40 $1000 Voice AI Competition Kickoff
$1000 Voice AI Competition Kickoff
Voiceflow
41 How to Build a Successful AI Agency | Voiceflow Panel Event
How to Build a Successful AI Agency | Voiceflow Panel Event
Voiceflow
42 AI Models are changing the way we build AI Agents | Humans Talking Agents Episode 3
AI Models are changing the way we build AI Agents | Humans Talking Agents Episode 3
Voiceflow
43 Faster Training, Better Intents | RAG Intent Recognition: Explained
Faster Training, Better Intents | RAG Intent Recognition: Explained
Voiceflow
44 Will voice AI kill call centers? | Humans Talking Agents Episode 4
Will voice AI kill call centers? | Humans Talking Agents Episode 4
Voiceflow
45 Build an AI agent in seconds — here's how.
Build an AI agent in seconds — here's how.
Voiceflow
46 Connecting multiple agents into an Agent Network with the new Agent step
Connecting multiple agents into an Agent Network with the new Agent step
Voiceflow
47 How will Vibe Coding affect software? | Humans Talking Agents Episode 5
How will Vibe Coding affect software? | Humans Talking Agents Episode 5
Voiceflow
48 Vibe coding: the end of coding as we know it
Vibe coding: the end of coding as we know it
Voiceflow
49 Vibe coding and resolution-based pricing — what will happen to AI companies' pricing models?
Vibe coding and resolution-based pricing — what will happen to AI companies' pricing models?
Voiceflow
50 Grow your AI agency: How to get new customers | Voiceflow Workshop Event
Grow your AI agency: How to get new customers | Voiceflow Workshop Event
Voiceflow
51 MCP is the key to an agentic internet | Humans Talking Agents Episode 6
MCP is the key to an agentic internet | Humans Talking Agents Episode 6
Voiceflow
52 MCP will change agent building forever with new standards for interactions
MCP will change agent building forever with new standards for interactions
Voiceflow
53 Review and improve your AI agent responses with call recording
Review and improve your AI agent responses with call recording
Voiceflow
54 4 tips to optimize your voice AI calls in Voiceflow
4 tips to optimize your voice AI calls in Voiceflow
Voiceflow
55 Launch AI agents even faster: new prompt generation feature
Launch AI agents even faster: new prompt generation feature
Voiceflow
56 Give your AI agents memory
Give your AI agents memory
Voiceflow
57 Can we build an AI Agent for a bank in 5 minutes?
Can we build an AI Agent for a bank in 5 minutes?
Voiceflow
58 Automate customer support tickets with AI (step-by-step Voiceflow tutorial)
Automate customer support tickets with AI (step-by-step Voiceflow tutorial)
Voiceflow
59 How to add custom ElevenLabs voices to Voiceflow
How to add custom ElevenLabs voices to Voiceflow
Voiceflow
60 Can we build an AI agent for Notion in 5 minutes?
Can we build an AI agent for Notion in 5 minutes?
Voiceflow

Learn how to use Voiceflow's new features, Structured Prompt Outputs and Variable Pathing, to build AI agents with more efficient variable management and specific JSON output formats from Open AI LLMs.

Key Takeaways
  1. Add a set step to the canvas logic
  2. Define a prompt with a specific output format
  3. Enable JSON output in the settings
  4. Define the output format with the required properties
  5. Use variable pathing to access properties of the output
  6. Implement conditional logic based on the output properties
💡 With Structured Prompt Outputs and Variable Pathing, users can build AI agents with more efficient variable management and specific JSON output formats, making it easier to integrate Open AI LLMs into their applications.

Related Reads

📰
Prompt Engineering Fails Quietly —  Prompt Regression Is Why
Learn to detect hidden regressions in prompt engineering to prevent silent failures in production and ensure reliable AI model performance
Towards Data Science
📰
Prompt Engineering: The Skill That Makes AI Work Better
Learn how to optimize AI performance with prompt engineering, a crucial skill for maximizing AI tool effectiveness
Dev.to · patil rushikesh
📰
5 prompt engineering techniques to get the best out of a legacy project
Learn 5 prompt engineering techniques to improve legacy project performance and why they matter for maintaining outdated codebases
Dev.to · Marco Coelho
📰
I Spent 30 Days Learning AI Prompt Engineering — Here’s What Actually Matters
Learn the key takeaways from a 30-day journey into AI prompt engineering and its practical applications
Medium · ChatGPT
Up next
I Built an AI Agent in 6 Minutes (No Code, No Developer)
HubSpot Marketing
Watch →