Skip to content
← All articles
Project introduction

Following financial news through AINewsTracker

Listen to the audio story

Turning financial news into data you can test · 2:22 · AI-written script, synthetic voices

Transcript

Ava: A headline lands, and within minutes it's mentioned in a thousand places. How do you turn that endless stream into something you can search, compare with market data, and test an idea on?

Andrew: That sounds like drinking from a fire hose. Is that the job here?

Ava: It is. Antoine's project, AINewsTracker, is a FastAPI service that turns financial news into records you can search, analyze and line up with prices.

Andrew: So what happens when a request comes in?

Ava: Picture an assembly line with four stations. A request reaches the news route and starts a collection job. The news service pulls the feed and creates article records. A metrics service reads the text. Then a repository stores the result in MongoDB.

Andrew: Four stations, one job each.

Ava: Right, and that's what keeps it tidy. Routes talk HTTP, models describe the data, services do the news work, and repositories keep database access in one place. Redis sits on top for caching.

Andrew: What does a stored article actually hold?

Ava: Its URL, publication date and text, next to the analysis fields. The documentation links four kinds of records: a news source, its articles, comments on an article, and companies. A source carries language and country. A company's ticker ties mentions together, so different spellings of a name still point to one business.

Andrew: And the analysis part?

Ava: The documented outputs are a sentiment label and score, detected tickers, a short summary and topic labels. Each answers a different question: which company is this about, what does it say, and how is it written.

Andrew: Then comes the trading side.

Ava: Yes. A strategy route hands backtest requests to a trading module. The labels become inputs to an experiment: pick a period, give the strategy the news that was available at that moment, and read what it reports back.

Andrew: So the source stays attached the whole way.

Ava: That's the neat part. A sentiment label traces back to its article and its publisher, and the date tells you when it was known. That makes every step easy to inspect.

Andrew: So what should a listener take away?

Ava: That you can build a clean path from raw feed to historical experiment, with each piece doing one job. The full write-up has the architecture diagram, and every box in it links to its source file.

AINewsTracker
Table of contents

AINewsTracker turns financial news into records that can be searched, analysed and compared with market data.

The project is a FastAPI service with separate routes for news, articles, companies, users, data seeding and trading strategies. Routes receive HTTP requests; services and repositories handle analysis and storage.

1. From a feed to an article

A news request starts a collection job. The news service retrieves feed content and creates article records. The metrics service analyses the text, and the article repository stores the result in MongoDB.

ResponsibilitySource file
Receive the requestapp/routers/news.py
Collect newsapp/core/services/news.py
Analyse the textapp/core/services/metrics.py
Store articlesapp/core/repo/article.py

Routes describe the API surface, models define stored data, services perform the news work, and repositories isolate database access. app/main.py also configures Redis caching.

AINewsTracker article processing flow

AlgoETS. (2026, November 5). Article processing flow. AINewsTracker repository, docs/diagrams/out/article_processing_flow.svg.

Original processing diagram[1].

2. Keep the source attached to the analysis

The documentation’s data overview[2] connects four records: a news source produces articles, comments belong to an article, and company records identify the businesses it mentions. Source metadata includes language and country. Articles retain their URL, publication date and text alongside analysis fields.

That separation makes a result easier to inspect. A sentiment label can be traced to an article and its publisher. A company’s ticker provides a way to connect mentions across articles without treating every spelling of its name as a different business.

3. What the text analysis produces

The documented NLP outputs include a sentiment label and score, ticker detections, a summary and topic labels. These fields answer different reading questions: which company is discussed, what the article says, and how its language is classified. The documentation’s sample records illustrate the model; the Pydantic classes define the actual fields.

I would inspect the original text beside those outputs before using them as strategy inputs. A summary can omit a condition, and a company mention does not establish that the article concerns a tradable event. The source URL and publication date therefore belong in the review path.

4. Analysis and backtesting

The project documentation describes sentiment analysis, topic classification, ticker allocation and article summarisation. The strategy route sends backtest requests to app/core/trading/strategy.py.

These labels are inputs to an experiment. They do not prove that a news item caused a market move, and a historical backtest does not guarantee future performance.

A useful review separates the data stages:

StageWhat to inspect
CollectionPublisher, source URL, publication time and retrieved text
Text analysisDetected companies, labels and summary against the original article
Strategy inputWhich observations were available at the point being tested
Backtest outputThe chosen period, assumptions and the strategy’s reported result

This review checks the data available to the strategy, its assumptions and its reported result. The article presents no measured prediction accuracy or trading returns.

5. Read the architecture

The complete GitDiagram includes route registrations, models, repositories, seeders, MongoDB, Redis and the trading branch. Each node links to its source file.

References

  1. AlgoETS. (). Original processing diagram. [Source file or documentation] github.com.
  2. documentation’s data overview. (). [Web page] algoets.github.io.
  3. Interactive GitDiagram. (). [Web page] gitdiagram.com.
  4. AlgoETS. (). AINewsTracker. [Software repository] github.com.
  5. Published documentation. (). [Web page] algoets.github.io.

References use this article’s publication date.