MarketPulse: Building an AI Marketing Analyst for Small Businesses

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle Submission 💸

This is a submission for the [Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle]

Marketing data can tell you what happened in a campaign. Understanding what to do next is a different challenge.

Small-business owners and freelance marketers may have campaign data showing their spending, clicks, conversions, and revenue, but turning those numbers into useful decisions can take time.

That’s the problem I wanted to address with MarketPulse.

MarketPulse is a marketing analytics application designed to help small-business owners, freelance marketers, and marketing teams understand campaign performance and make better-informed marketing decisions.

I built it to bring several tools into one place:

The goal is not to replace a marketer’s judgment. It is to make marketing data easier to understand and give users a clearer starting point for their next decisions.

The video walks through the application and demonstrates its main features.

I built MarketPulse using Python and Streamlit.

The application combines deterministic calculations with language-model assistance. Python handles the key calculations, while the AI assistant uses the available marketing context to explain results and suggest possible next steps.

During testing, I realized that an AI-generated explanation could sound convincing even when a numerical calculation was wrong. I strengthened the validation around combined channel-share calculations and changed the application to pass calculated results to the model.

I also improved the instructions given to the assistant to reduce unsupported metrics and prevent it from confusing return on ad spend (ROAS) with profit.

That experience taught me an important lesson: * a useful AI product needs more than a language model. It also needs reliable calculations, validation, and clear limits on what the model can claim. *

I tested the application with marketing campaign data and considered its practical usefulness through my experience running Ghifty’s Bowl, my fruit-salad business. MarketPulse is still being developed and validated.

MarketPulse uses the open-weight openai/gpt-oss-20b model to power its AI Marketing Assistant.

I chose this approach because open-weight models give developers another way to build AI-powered products and experiment with how models can support specific tasks.

In MarketPulse, the model helps interpret campaign metrics, but the application’s calculations and validation logic remain in Python. This allows me to build product-specific safeguards around the AI rather than relying on generated text alone.

The model currently runs through Groq’s hosted inference API. It does not run locally in MarketPulse, and requests to the AI assistant are processed through that hosted service.

For me, open innovation matters because it gives builders more choices and room to experiment. It allows developers to explore different models and deployment approaches as their projects evolve.

I want MarketPulse to demonstrate how AI can become part of a practical tool when combined with data analysis, thoughtful engineering, and an understanding of the people it is meant to serve.

I built and tested the application using Python and AI-assisted development workflows. Rather than include a session that I cannot verify or share, I’m focusing this submission on the working application, its code, and the decisions I made while developing it.

I have not yet confirmed an external user or paying customer for MarketPulse.

My next step is to get the application in front of small-business owners and marketers, collect feedback, and learn which features solve their most important problems.

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle.