Building a Financial Data Marketplace with AI Agents

I never set out to build a data marketplace.

I'm a data scientist from France who has spent way too many years wrangling financial datasets that should have been easy but never were. If you've ever tried to do something simple — like get a clean time series of Treasury yields alongside earnings dates and sector classifications — you know what I mean. The data exists. It's technically "public." And it's spread across seventeen different government websites, three file formats, and at least one system that still thinks it's 1997.

This is the origin story of kibble.shop. It's a little embarrassing, a little nerdy, and entirely true.

The frustration that started it all

About a year ago, I was working on a personal project — nothing fancy, just trying to build a simple dashboard that combined macro indicators with equity market data. The kind of thing you'd think would take a weekend.

It took three weeks.

Not because the analysis was hard. Because the data was hard. Every source had its own format, its own quirks, its own way of handling missing values (my personal favorite: one API returned nulls as the string "N/A", another as -999, and a third just... skipped the row entirely). Joining anything across sources was an exercise in creative key matching and prayer.

And I kept thinking: someone must have solved this already. Surely there's a place where financial data just... works. Where it's clean, documented, queryable, and you can trust it enough to build on top of.

There wasn't. Or at least, not one that didn't cost $20,000/year and require an enterprise sales call.

The side project that grew teeth

So I started building it myself. Just for me, at first. A set of Python scripts that pulled data from various free sources, cleaned it up, standardized the formats, and dumped everything into a local database with some basic quality checks.

It was ugly. It worked. And it kept growing.

Every time I needed a new dataset, I'd add another ingestion pipeline. Options flow data from the SEC. Yield curves from Treasury.gov. Earnings calendars. Economic indicators. Each one got the same treatment: pull, clean, validate, standardize.

After a few months, I had something that was genuinely useful — not just for my original project, but for a dozen other things I wanted to explore. And I started wondering: if this is useful to me, maybe it's useful to other people too?

That's when the side project became kibble.shop.

Why "kibble"?

People ask this a lot. The honest answer: data is the food that feeds models, analyses, dashboards, and decisions. Good data is good kibble. Bad data is... well, you know what happens when you feed something garbage.

Also, I just liked the name. It's memorable, it's a little weird, and it doesn't take itself too seriously. In a world of "Quantum Data Analytics Platforms" and "Enterprise Insight Hubs," I wanted something that felt human.

What we've built

As of today, kibble.shop has 185+ data products spanning:

• Treasury & fixed income — yield curves, spreads, rate expectations
• SEC filings — insider transactions, institutional holdings, corporate events
• Macro indicators — employment, inflation, GDP, leading indicators
• Market structure — options flow, volatility surfaces, sector rotations
• Derived products — things we've built on top of the raw data, like anomaly scores, composite indicators, and cross-dataset joins that would take you weeks to build yourself

Every data product has a contract — a machine-readable definition of what it contains, how fresh it is, and what quality checks it passes before publishing. This isn't just metadata for show. If a quality check fails, the data doesn't ship. Period.

The part where AI agents enter the picture

Here's where it gets interesting. kibble.shop isn't just a website with some CSVs. We're building it to be as useful to AI agents as it is to humans.

Think about it: if you're building an AI assistant that can answer financial questions, or a trading bot that needs macro context, or an agent that monitors insider activity — where does it get its data? Scraping random websites? Calling fifteen different APIs with fifteen different auth methods?

We want kibble.shop to be the answer to that question. Clean data, consistent formats, proper APIs, and documentation that both humans and machines can parse. One place to get the financial data you need, whether you're a person with a Jupyter notebook or an agent with an API key.

What I've learned so far

A few things that have surprised me along the way:

Data quality is the whole game. Not the analysis, not the visualization, not the UI. If the data is wrong, nothing downstream matters. I spend more time on validation and quality checks than on anything else, and it's worth every minute.

People are desperate for accessible financial data. The response from early users has been overwhelming. Turns out there's a huge gap between "data exists somewhere on the internet" and "data I can actually use in my work without wanting to throw my laptop out the window."

Building in public is terrifying and great. Sharing what you're building before it's perfect feels vulnerable. But the feedback loop is incredible. Users catch things I miss, suggest products I hadn't thought of, and keep me honest about what actually matters.

The "last mile" of data is the hardest mile. Getting data from source to a clean, reliable product involves dozens of decisions: how to handle gaps, how to normalize across sources, how to version changes, how to communicate freshness. It's unsexy work that makes or breaks the product.

What's next

We're still in early access — everything on kibble.shop is free right now. The focus is on building the best possible data products and getting feedback from real users. More datasets are coming every week.

If you work with financial data — whether you're a quant, an analyst, a developer building fintech tools, or just someone who wants clean macro data without the headache — come check it out. Poke around. Break things. Tell me what's missing.

This started as a frustrated data scientist's side project. I'd love for it to become the place where people actually enjoy working with financial data. Ambitious? Maybe. But someone's gotta feed the models.

Sign up for free and explore 185+ financial data products. I'd love to hear what you think — and what you'd want us to build next.