Explained: The Voiceflow Knowledge Base (Retrieval Augmented Generation)
This video explains how Retrieval Augmented Generation works and goes through a deep dive of what happens after you upload a document to the Voiceflow knowledge base.
You'll learn how chunks work, what a vector database does, and how to optimize responses from the Voiceflow Knowledge Base.
Let us know if you have questions in the comments below!
***
The collaborative platform to build AI agents. Use Voiceflow to design, test, and launch chat or voice AI agents - together, faster, at scale.
Join our Discord community
👾 / discord
Kickstart your next project with our templates
🚀 www.voiceflow....
Our Links 🔗
👉 Start building today: www.voiceflow.com
👉 Subscribe: bit.ly/3am22nf
👉 Twitter: bit.ly/2xrXZqV
👉 LinkedIn: / voiceflowhq
👉 Publication: www.voiceflow....
Пікірлер: 19
Another amazing video! Thank you One of the reasons I chose Voiceflow is the explanations you provide, Unequivocally this is the factor that affects the choice! Don't stop uploading videos and explaining this is what makes you a level above everyone else.
@Voiceflow
8 ай бұрын
Thank you for the feedback! We’ll keep making these concepts easier to understand and apply. Let us know what else we should cover :)
@user-rj6cw4sk9l
8 ай бұрын
I like to see more explain about creating costume functions in voiceflow for my personal bot , full tutorial. Thank you for the response! @@Voiceflow
@user-rj6cw4sk9l
8 ай бұрын
answer about Tokens Q , such as - If I use Asisstant API according to your video with Voiceflow , How the tokens is billed? according to the OpenAI or Voiceflow ? @@Voiceflow
🎯 Key Takeaways for quick navigation: 00:00 📚 *Introduction à la génération augmentée par récupération (Retrieval Augmented Generation - RAG)* - Présentation de la fonctionnalité RAG pour les assistants AI et importance de comprendre son fonctionnement. 00:28 💡 *Explication du fonctionnement de la base de connaissances Voiceflow* - La base de connaissances Voiceflow permet d'uploader des documents et de concevoir des assistants AI pour accéder à ces documents et répondre aux questions. - Utilisation de la couche de base de connaissances de l'IA et intégration de documents supplémentaires pour répondre aux questions. 01:24 🖥️ *Processus de téléchargement des documents dans Voiceflow* - Décomposition des documents téléchargés en sections pour traitement. - Illustration du processus de traitement des documents téléchargés et de leur préparation pour l'utilisation. 03:16 🗃️ *Stockage et utilisation des morceaux de documents dans une base de données vectorielle* - Conversion des sections de documents en vecteurs dans une base de données vectorielle. - Explication de la façon dont les questions des utilisateurs sont traitées en comparant avec les vecteurs de la base de données. 05:19 🤔 *Processus de réponse aux questions par le modèle d'IA* - Sélection et envoi des sections pertinentes au modèle d'IA pour générer une réponse. - Présentation de la manière dont les sections sont choisies et utilisées pour répondre aux questions des utilisateurs. 06:32 🔍 *Démonstration pratique de la fonctionnalité de base de connaissances* - Exemple concret d'utilisation de la base de connaissances pour répondre à une question spécifique. - Visualisation du processus dans l'interface utilisateur de Voiceflow. 09:07 ⚙️ *Réglages et optimisations des modèles d'IA dans Voiceflow* - Discussion sur le choix des modèles d'IA, les réglages de température et la gestion des tokens. - Conseils pour optimiser les réponses en fonction des besoins spécifiques. 11:38 🛠️ *Conseils pour améliorer la qualité des réponses et débogage* - Stratégies pour améliorer la précision des réponses en affinant les documents sources. - Utilisation du mode preview pour identifier et rectifier les sources d'erreurs dans les réponses. 13:20 🌐 *Utilisation des API Voiceflow pour la gestion de la base de connaissances* - Présentation des API pour télécharger, supprimer ou remplacer des documents dans la base de connaissances. - Exemples d'applications pratiques des API pour les développeurs et les entreprises. Made with HARPA AI
Nice and clear presentation, good job💪👏
@Voiceflow
8 ай бұрын
Glad to hear!
I have tried asking this question if a few places: how does the knowledge base chunk information from spreadsheets? When I upload a small sample spreadsheet to the KB, it chunks the information in a way that doesn't follow the structure of the spreadsheet. For example, it will chunk an entire column of data together rather than looking at the row as the individual data set that needs to be returned. So when I ask the bot to give me all of the information about data point, lets call it "Rocky", it does a fine job returning all of the data about "Rocky", but when I ask it to return a specific piece of data about "Rocky" such as "Rocky's phone number", it will return "John's" phone number instead. When you look at the chunks, it shows that Rocky and John's phone numbers are being chunked together. Why can't it distinguish things that are clearly defined on a spreadsheet?
One question, please. So if my customer doesn't want to share their documents with openAI. So I upload the in VF. Then use gpt3 5 for example. But VF passes the chunks to OpenAI, which violates my customer requirements. Who knows what openAI is doing with my chunks? Maybe they store them as additional KB? Or maybe they will in the future? Am I missing something?
I have one question, if you are able to answer me. I have set the URL and one docs in Knowledge Base , and when I test the AI in designer mode (Response AI-->Preview), it gives me an answer in Serbian (correct), and when I play it on RUN or production, it answers in English(not correct). otherwise url and docs are in Serbian language. Thanks and you are doing a good job
A question that arises from the video - Now after the update of KB, is there a difference between building a bot with Assistant API or with Voiceflow and putting files in KB? In terms of the understanding of the files and the analysis of the information (assuming that both in the Assistant and in the KB of Voiceflow I put the same model for example GPT 4)
@Voiceflow
8 ай бұрын
This video explains it in depth if you haven’t watched it yet Assistants API vs Voiceflow: Build 10x faster with the same models kzread.info/dash/bejne/ZI6ms8iTotPAgdo.html
Exceptional explnations! Cristal Clear
@Voiceflow
8 ай бұрын
Glad it was helpful!
Daniel. At the beginning of the video, you mentioned that you would explain how to retrieve specific documents. I listened to the entire video, but you didn't explain that...
@Voiceflow
8 ай бұрын
Good callout! To do this you can use the Tags API. This allows you to add tags to specific documents then, using the Query API you can outline which documents to include or exclude. We’re adding this functionality to the UI at the end of January. But you can use it today via the APIs developer.voiceflow.com/reference/post_v3alpha-knowledge-base-tags
@Aidev7876
8 ай бұрын
@@Voiceflow thanks Daniel. I'll wait for JANUARY for it to be available from the UI. Appreciated.
Low key I see that Fenty Beauty flex.... I see you playa...
@Voiceflow
7 ай бұрын
Flossy, glossy 👀