Image by Editor
#Introduction
It’s a myth that deploying artificial intelligence (AI) projects requires months. The truth is, you can deploy an AI analyst that can answer complex business questions from your own Structured Query Language (SQL) database in minutes if you know how to connect the right large language model (LLM) to your data source successfully.
In this article, I’m going to break down how to deploy an AI analyst with Bag of Words, an innovative AI data layer technology. You will learn practical, step-by-step processes that focus on SQL databases and LLMs. Along the way, we will cover common deployment struggles and ethical considerations every professional should know.
#Understanding Bag of Words
Bag of Words is an AI data layer platform that connects any LLM to almost any data source, including SQL databases like PostgreSQL, MySQL, Snowflake, and more. It helps you build conversational AI analysts on your data with these key features:
- It allows direct connection to your existing data infrastructure
- It controls which tables and views the AI can access
- It improves your data context with metadata from tools like Tableau or dbt
- It manages user access and permissions securely
- It is designed for fast, trustworthy, and explainable insights
This approach simply means that users can “ask once, improve, and get results you can explain,” all without huge engineering expenses.
Image by Editor (click to enlarge)
#Deploying an AI Analyst
Many organizations struggle to unlock the full potential of their data, despite having powerful tools. The problem is mostly integration, which is complex, and there is no clear method of integration. AI analysts powered by LLMs transform raw data into insights through natural language queries, but accurately connecting these models to backend data is crucial.
The good news is that Bag of Words has made it possible to connect your SQL databases and LLMs without having issues with endless custom code. This lowers barriers and speeds deployment from weeks or months to minutes, empowering both data teams and business users.
#Deploying an AI Analyst with Bag of Words
Follow these technical steps to get an AI analyst up and running rapidly in Docker.
//Step 1: Preparing Your SQL Database
- Ensure that Docker is installed on your machine and set up correctly before running the code below.
- Then run the following command:
docker run --pull always -d -p 3000:3000 bagofwords/bagofwords- You will need to sign up if you’re new:
http://localhost:3000/users/sign-up.
Image by Author
Follow the steps to complete the onboarding flow to set up your AI analyst.
- Make sure you have your connection credentials for your SQL database (host, port, username, password).
- Click New Report. Then select any database of your choice. For this article, I will go with PostgreSQL.
Image by Author
- Create your database and populate it. I recommend Supabase for the demo. You can use any one of your choice. Also, ensure your database is accessible from the network where you’ll deploy Bag of Words.
Image by Author
- Know which schemas, tables, and views have the data you want the AI analyst to query.
- Next is to give context to your analysis.
Image by Author
This is where you need to give the AI instructions on how you want the data to be managed, and you can connect with Tableau, dbt, Dataform, and your AGENTS.md files in Git.
You can also set up a conversation where, with a click of a button, you have your answer ready with all the information you need.
Image by Author
You can also set up and rerun your report. The report on your data becomes an autopilot.
Image by Author
//Step 2: Testing and Refining Queries
- Interact with the AI analyst via the Bag of Words interface.
- Start with simple natural language queries like “What were total sales last quarter?” or “Show top products by revenue.”
- Refine prompts and instructions based on initial results to improve accuracy and relevance.
- Use debugging tools to trace how the LLM interprets SQL and adjust metadata if needed.
//Step 3: Deploying and Scaling
- Integrate the AI analyst into your business applications or reporting tools through APIs or user interface (UI) embedding.
- Monitor usage metrics and query performance to identify bottlenecks.
- Expand database access or model configurations iteratively as adoption grows.
#Challenges and Solutions
Here are some roadblocks you may face when when deploying AI analysts (and how Bag of Words can help):
| Model | Train Acc | Val Acc | Gap | Overfitting Risk |
|---|---|---|---|---|
| Logistic Regression | 91.2% | 92.1% | -0.9% | Low (negative gap) |
| Classification Tree | 98.5% | 97.3% | 1.2% | Low |
| Neural Network (5 nodes) | 90.7% | 89.8% | 0.9% | Low |
| Neural Network (10 nodes) | 95.1% | 88.2% | 6.9% | High – Reject this |
| Neural Network (14 nodes) | 99.3% | 85.4% | 13.9% | Very High – Reject this |
#Wrapping Up
Deploying an AI analyst in minutes by connecting any LLM to your SQL database is not just possible; it is becoming expected in today’s data-driven world. Bag of Words offers an accessible, flexible, and secure way to rapidly turn your data into interactive, AI-powered insights. By following the outlined steps, both data professionals and business users can unlock new levels of productivity and clarity in decision-making.
If you’ve been struggling to deploy AI projects effectively, now is the time to demystify the process, harness new tools, and build your AI analyst with confidence.
Shittu Olumide is a software engineer and technical writer passionate about leveraging cutting-edge technologies to craft compelling narratives, with a keen eye for detail and a knack for simplifying complex concepts. You can also find Shittu on Twitter.
