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