Inteldo

Inspiration

Business questions rarely have their answer in one place.

A simple question about growth can involve analytics, advertising, payments, product usage, SEO, and external research. We found ourselves spending a lot of time moving between different tools, pulling data out, comparing numbers, and trying to piece everything together before we could even start answering the actual question.

That led to a simple idea: instead of making people collect all the data first, what if the research process could handle that work itself?

What it does

Inteldo investigates business questions across connected data sources and external sources.

A question is broken into focused investigations, with specialist agents handling different parts of the research. One agent might look at revenue, another at traffic or product usage, while another researches external information.

The findings are then brought together into one answer, with supporting sources and citations so the research can be checked.

Inteldo can work with sources such as Stripe, Google Ads, Google Analytics, Google Search Console, Ahrefs, and PostHog.

The goal is simple: turn scattered business data into clear answers.

How I built it

Inteldo uses a multi-agent architecture where different agents are responsible for different types of research.

When a question comes in, the system determines which areas need to be investigated and sends those tasks to the relevant agents. They query the connected sources, collect the relevant information, and return their findings.

A synthesis layer then brings those findings together into a single response while preserving the supporting sources.

The agents are also designed around specific research roles rather than having one general-purpose agent try to do everything. This helps keep investigations focused and makes it easier to understand where different parts of an answer came from.

Challenges I ran into

One of the biggest challenges was dealing with the fact that different sources describe the same business in different ways.

Metrics can use different definitions, time periods, or levels of detail. Two sources can also appear to disagree even when both are technically correct.

Another challenge was deciding how much work to delegate to separate agents. Splitting everything into smaller tasks can create unnecessary complexity, while giving one agent too much responsibility makes the research harder to inspect.

We also had to make sure that the final answer did not lose the evidence collected during the individual investigations. A useful answer is much more valuable when you can see where the information came from.

Accomplishments that I'm proud of

I'm proud that Inteldo can take a question that would normally require manually checking several different systems and turn it into one coordinated research workflow.

The multi-agent approach also makes it possible to investigate different parts of a question in parallel instead of treating the research as one long sequence of steps.

Most importantly, we built the system around traceability. The findings are accompanied by supporting sources and citations rather than simply presenting a conclusion without context.

What I learned

I learned that building a multi-agent system is less about creating more agents and more about deciding what each agent should actually be responsible for.

Good delegation matters. So does knowing when information from different sources can be combined and when apparent differences need to be investigated instead of averaged together.

I also learned that generating an answer is only part of the problem. For research workflows, being able to understand and verify how an answer was produced is just as important.

What's next for Inteldo

We're continuing to improve the research workflow, particularly how questions are broken down, how agents coordinate their investigations, and how conflicting information is handled.

We're also working on expanding the range of connected sources so more business questions can be investigated without requiring people to manually gather data first.

The longer-term goal is to make Inteldo useful for the messy questions that don't fit neatly inside a single dashboard or tool.