AI Build Week – Day 2: Vibes-Based Search with Vectors 🔍

[sarahs]

Welcome back to Day 2 of ai-build-week! Today we’re building an AI-powered movie search app that goes beyond keywords — surfacing the perfect MCU pick based on vibes alone.

AI Build Week – Day 2: Build a Vector-Powered AI Search App

Live at: 2025-07-08T17:00:00Z→2025-07-08T18:00:00Z — right here in the forum
Say hey if you’re tuning in! We’re chatting all day and answering questions.

---

Today’s session is all about RAG (retrieval-augmented generation) — but with a twist.
@angeliklaboy will show you how to upload documents to Retool Vectors, feed them into an LLM, and create an emotionally intelligent search experience. Think: movies that make you cry, laugh, or cheer based on real reviews.

---

What’s you'll learn:

• Upload + embed documents with Retool Vectors
• Design prompts for emotional or thematic search
• Build a full app + workflow for mood-based MCU recs

---

Get involved:
Download today’s resources
Poll: What’s your builder mode today? (Vote below!)
Got a cool vibes-based search idea? Share it in the thread
Sneak peek:
We’re giving away a few of our limited-run Retool baseball caps at the end of the week (Community Showcase folks know how good they are). Want one? Join the convo!

---

Missed Day 1? Watch the Prompt Arena replay here

← Back to the full AI Build Week schedule

[sarahs]

Poll: What’s your builder mode today?

(Choose the one that best matches your vibe right now.)

• Precision Mode – I know exactly what I’m building.
• Experiment Mode – I’m here to test things out and see what works.
• Speed Run Mode – I’m building fast and breaking things.
• Straight Vibing – Just here to watch and learn.

[sarahs]

Day 2 Q&A Recap

Thanks to everyone who joined us live for Day 2 of AI Build Week — focused on vectors and retrieval-augmented generation (RAG)! We got a ton of great questions from builders across experience levels, so here’s a quick recap:

---

Vector Setup & Data Management

Q: Can I upload multiple PDFs into a vector?
Yes! You can upload several PDFs into a single vector — they’ll all be chunked and indexed. Use filter labels or metadata for better control at query time.

Q: Can I filter vector results by specific documents?
Yep — just apply filter labels when uploading your files, and use them in your vector query to narrow the scope.

Q: Is there a difference between uploading one large file vs. multiple smaller ones?
Smaller, topic-focused files tend to perform better depending on your chunking and embedding setup. Try both and compare.

Q: How much data can I store in a Retool vector DB?
Retool currently supports up to 120 million embedding tokens per day — plenty for most use cases.

Q: How should I clean or prep my PDFs before uploading?
Stick to mostly text, focus on one topic per doc, and avoid image-heavy files. For transcripts, add speaker tags or break them into logical sections.

---

Automation & Updating Vectors

Q: Can I build a RAG app using data from an API?
Yes! Use a workflow to fetch, parse, and add data to a vector — great for support bots or embedding product info.

Q: Can I update vectors automatically from a URL or Drive folder?
Not natively, but you can create a scheduled workflow to check for updates and upsert new content.

---

Integration & Triggering

Q: Can I trigger a workflow from Slack or an external POST request?
Definitely. Workflows support webhooks — so you can trigger them from Slack, Zapier, or any tool that can send a POST.

Q: I already have a vector DB in Pinecone. Can I use that?
Yes! Connect Pinecone (or other external vector DBs) as a resource and query it directly in your apps or workflows.

---

Architecture & Best Practices

Q: Why build this as a workflow instead of using an agent?
Workflows are great for structured, linear logic like searching and returning results. Agents are better for open-ended decision-making — we’ll cover those more in Day 5.

Q: My long transcripts aren't returning great results. Any tips?
Try chunking by paragraph or speaker and add metadata. Cleaner, more focused chunks improve vector search quality.

Q: I built a chatbot using a syllabus file, but my prompt is too long. Any way to fix that or make it more efficient?
A: Instead of pasting the entire file into the prompt, upload it to a vector DB and retrieve only the most relevant chunks. This reduces token usage and improves prompt quality.

Q: I'm seeing an "undefined" error during the build-along. What's going wrong?
This usually means one or more blocks in your workflow haven’t been run yet. Be sure to click the Play button in the top right corner to run all previous blocks. You’ll also want to set a valid JSON test parameter so your data is properly initialized.

---

Let us know your biggest takeaway from Day 2 or drop a question if you’re stuck — we’re here for it.
See you for Day 3: Behind the Build – How We Made RetoolGPT