My Library

Day 5: Real Product Demo & What's Next

What You've Built

Over the last four days, you built something pretty cool. Even though it might not feel that complex - and that's kind of the secret when it comes to building with AI - it's just software development. It's just a new type of API.

You've learned:

  • Day 1: How to call an LLM API (text in, text out)
  • Day 2: System prompts to control AI behavior
  • Day 3: RAG with a knowledge base
  • Day 4: Adding real data with YouTube transcripts

You've also been exposed to concepts like vector databases, retrieval augmented generation, and context windows. Now you have areas you can explore to extend what you've already built.

Real Product Demo: TikTok Creator Finder

Let me show you a product I built for an AI startup. This is the original beta version that was used by Roc Nation and Universal when they wanted to find TikTok creators to help promote songs.

How It Works

The product uses the same concepts you learned - just at a larger scale:

  1. Data Ingestion: Millions of pieces of data about TikTok creators
  2. Vector Database: Creator profiles, video content, engagement metrics stored as vectors
  3. User Input: Form fields extract context (genre, budget, target countries, content type)
  4. RAG at Scale: SQL queries + vector search + LLM reasoning
  5. Results: Matched creators with pricing and reach estimates

Example Query

Artist: Bad Bunny
Genre: Reggaeton
Budget: $1,200 per influencer
Target Countries: Mexico, Colombia
Content Type: Lip sync pages

The system then:

  • Runs complex SQL queries
  • Searches the vector database for relevant creators
  • Extracts and processes information
  • Returns a curated list of creators with pricing

The Key Insight

This is not so different from what you've built. It's just expanding on the concepts you already learned:

  • Instead of a JSON file → Vector database with millions of records
  • Instead of simple keyword matching → Semantic search with embeddings
  • Instead of one transcript → Thousands of creator profiles
  • Same core pattern: Retrieve relevant data → Inject into prompt → Generate response

What Companies Are Building

I've also helped build:

  • Supplier Finder: For Fortune 500/100 companies to find suppliers through complex internet search
  • Internal Knowledge Bases: RAG systems for company documentation
  • AI-powered Search: Semantic search replacing traditional keyword search

This stuff is hard to hire for. I designed the AI engineer interviewing process for one company, and you would not believe how difficult it was to find people that knew:

  • Retrieval augmented generation
  • How to build things with agents
  • The pitfalls and edge cases
  • How to observe and test AI systems
  • How to make sure these things don't fail in production

The Opportunity

There's just not a lot of people that know this stuff right now.

You may think the market is saturated, but in reality:

  • Very few companies are building AI products the right way
  • Even fewer developers know this stuff at all
  • This knowledge will become table stakes - like knowing React or AWS

In the near future, knowing how to:

  • Build agents
  • Structure a vector database
  • Add observability and evaluations to AI systems

...will be expected, just like knowing cloud providers is expected today.

Two Types of Developers

There are developers working on: "How do I get better at prompting tools and write more code?"

And there are developers thinking: "How do I make the infrastructure and products that are gonna sell and make our company money?"

Knowing how to build AI systems is how you level up.

What To Do Next

Share What You've Learned

  • Share this with your company and teammates
  • Tell them: "Here's how RAG can work for us"
  • Maybe you don't even need a vector database - maybe the naive approach works for your use case
  • Do a hackathon at work to introduce and socialize these ideas

Extend Your Project

  • Add more creators to your advisory board
  • Try a different niche entirely (business, fitness, cooking)
  • Implement basic chunking for large transcripts
  • Experiment with a local vector database like Chroma

Keep Building

The best way to learn is by doing. Take what you've built and make it your own.

Ready To Go Deeper?

If you enjoyed this course and want to take your AI engineering skills to the next level, apply to Parsity's AI Developer Cohort.

This is a hands-on program where you'll:

  • Build production-ready AI systems with real-world data
  • Learn vector databases, embeddings, and semantic search
  • Understand observability and evaluation for AI systems
  • Work directly with senior engineers who build this stuff daily
  • Get the skills that AI startups are desperately hiring for

It's not just tutorials - it's humans in the loop, guest speakers, and seeing how people actually building these systems work through real problems.

Apply to the AI Developer Cohort →

Get In Touch

If you have feedback, want to share what you've built, or are confused about any aspect of this:

I'd genuinely like to see what you're building. If you've extended this, figured out how to use a vector database, or have questions - reach out!

Final Thought

Whether or not you go any further, keep this in mind: You have a really interesting skillset with just this tiny bit of information that a lot of developers don't know about.

This time has a lot of leverage and opportunity - even though there's fear and uncertainty at the same time.

Thanks for taking the time to do this program. Now go build something!