About

Polly, a finance AI engine

We call Polly an AI agent, because that is the word everyone uses now. What sits behind the name is closer to a large engine. It is an orchestration layer, a harness that takes a question, picks a model for the job, and arms it with tools. Some of those tools search the web. Most of them are ours. We build them and we maintain them, and that is where most of the work goes: how precisely the engine runs comes down to how good its tools are.

What it is good for

What an engine buys you first is attention. Nobody can read a few thousand stories a day, and almost none of them deserve to be read. Machines are good at exactly that job: reading all of it, discarding most of it, and leaving a short list of what actually moved.

The second thing is memory. Markets return to the same handful of arguments for years, and people forget where the last chapter left off. The archive does not. A story arrives filed next to the ones it belongs with, which is usually the fastest way to tell whether something is new or merely loud.

The third is the tedious part. Most of the work in a question about a company is fetching: the filing, the last four quarters, what the analysts did, who was buying. Polly does the fetching and shows what it used. The judgment is still yours. It just arrives sooner.

Limitations

An engine like this has a ceiling, and the ceiling is what it can read. Where our coverage is thin the answers are thin too, and nothing about them says so. The machinery cannot tell a quiet day from a day we missed.

Most of the work underneath is judgment at a scale no desk could staff: what a story is about, whether it matters, which company it really touches. Judgment at that scale is statistical. It is right far more often than not, and when it is wrong it is wrong quietly.

And a signal is not a cause. We can show what happened and what was published around it; why a price moved is a story told afterwards, and usually with more confidence than anyone has earned.

Our goals

The long goal is easy to say and hard to build: give one reader the coverage that used to take a desk. Not a verdict and not a newsletter, but the same record an institution works from, organized well enough that a single person can hold it.

Most of the way there runs through sources and depth. Closer to the primary record and further from the vendors who resell it. More of what we already hold made readable, since an archive is worth more read than bought. Fewer feeds, better ones.

And precision comes before scale. The limitations above are work items, not disclaimers we plan to keep. What we watch is narrow: how often a reader has to go check something themselves, and how easily they can when they do.

A few more things

Stockpolly doesn’t give investment advice. It won’t tell you what to buy or sell or where a price is going, and if you ask Polly, it will decline. Organizing the information is our job; the decision is yours.