The tech stack for real-time recommendation systems

The hidden complexity of AI integration: A personal perspective. The future of database querying: Tabular semantic search.

Paul Iusztin

Feb 08, 2025

The future of database querying: Tabular semantic search
The hidden complexity of AI integration: A personal perspective
The tech stack for real-time recommendation systems

The future of database querying: Tabular semantic search

Forget about text-to-SQL for querying databases.

There’s a 10x improvement that changes the game:

𝗧𝗮𝗯𝘂𝗹𝗮𝗿 𝘀𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝘀𝗲𝗮𝗿𝗰𝗵 𝘂𝘀𝗶𝗻𝗴 𝗺𝘂𝗹𝘁𝗶-𝗮𝘁𝘁𝗿𝗶𝗯𝘂𝘁𝗲 𝘃𝗲𝗰𝘁𝗼𝗿 𝗶𝗻𝗱𝗲𝘅𝗲𝘀.

Why is it so good?

It integrates structured and unstructured data into one search system.

Here’s how we implemented this for Amazon products using Superlinked as our tabular semantic search engine and MongoDB as our vector database:

1. Tabular semantic search

We search across attributes like descriptions, ratings, and prices based on similarity but with a twist: weighting.

results = app.query(     …,     description_weight=1.0,     reivew_weight=0.5,     price_weight=0.7,     query=”books about LLMs and RAG that have good reviews”  )

This query finds products with a description similar to the user's query.

The real power comes from the weights applied to different variables like:

description_space (text)
reivew_weight (number)
price_weight (number)

Adjust description_weight for relevance or tweak price_weight for price sensitivity.

And if you're using natural language queries, the system applies these weights automatically.

2. Semantic query with with_vector()

This extends the search beyond keywords to find products similar to a specific item.

similar_items_query = semantic_query.with_vector(sl.Param("product_id"))

This is ideal for "more like this" functionality or recommendations.

But here's where things get interesting.

Superlinked lets users enter natural language queries, which the system decodes into structured search parameters and embeddings.

It’s like text-to-SQL but more powerful.

For example, for query: “Find books under $100 with a rating over 4.”

Superlinked decodes it into:

query_description = books
query_price = 100
review_rating = 4

No need to manually structure this query - it’s done seamlessly by the system.

(As the query is declarative, it dramatically reduces error possibilities relative to text-to-SQL)

So, what's the benefit of assigning weights to embeddings?

Weights allow you to fine-tune relevance without needing to re-embed or re-index your entire dataset.

This can save a massive amount of time.

(especially when iterating and experimenting with search configurations)

For example:

Increase the description_weight for more focus on product descriptions.
Adjust the price_weight if price is a key factor.

Without these weight adjustments, experimentation would be painfully slow due to re-embedding and re-indexing.

Forget rigid SQL syntax.

This is the future of querying:

Flexible
Intuitive
Powerful

Want to learn more about how to implement this?

Consider reading our free step-by-step course on building an Amazon e-commerce search app using tabular semantic search and natural language queries:

GO TO COURSE

The hidden complexity of AI integration: A personal perspective (Affiliate)

As an engineer and content creator deeply involved with Generative AI, I've encountered an unexpected challenge: effectively integrating AI into daily workflows.

This isn't about complex LLM architectures or model optimization—it's about the mundane yet critical tasks that consume valuable time.

Common workflow challenges:

Context-aware email composition using Notion
Project timeline automation for team collaboration
Converting research notes into social media content
Developing course landing pages with hybrid code-template approaches
Transforming research into long-form content

Despite extensive AI experience, these tasks still demand significant time investment. While using ChatGPT is straightforward, developing sustainable, automated workflow systems presents a more nuanced challenge.

Introducing: How To Scale A Business With AI & Agentic Workflows

Reviews from the **“How To Scale A Business With AI & Agentic Workflows - Foundations”** live course - **Enroll here**.

An intensive 5-week, live cohort course designed to teach you how to:

Build AI assistants that boost your efficiency.
Automate your most time-consuming workflows.
Implement agentic systems that deliver ROI for your business.

And it already has an outstanding 5.0 🌟 rating from 42 reviews.

The transformation is real.

Past students have scaled their business operations, reduced workflow friction, and found more time to focus on high-impact work.

And with Sara Davison leading the way, you’re learning from a true expert with over 8,000 hours of hands-on AI implementation.

The next cohort starts March 4th.
Get $100 off with code: PAUL
100% company reimbursement eligible

If you want to learn how to build systems and scale your daily processes using AI, consider enrolling:

GO TO COURSE