FSS
Financial Sentiment Signals
Backtest Console · v2.0
71.4% Accuracy
36 Articles/day
5 Sources

AI-powered sentiment analysis. Backtest trading signals, track prediction accuracy, and uncover actionable market insights.

Built with AWS Lambda · SageMaker · DynamoDB
Welcome back
Sign in to your FSS account
Don't have an account?
Create account
Start analyzing financial sentiment signals
Already have an account?
Verify your email
We sent a 6-digit code to your email
Reset password
Enter your email and we'll send you a reset code
Remember your password?
Set new password
Enter the code we sent and your new password
FSS Financial Sentiment Signals
Backtest Report

General Tech Sentiment

Market-wide · loading…
1-day horizon
Trading Days
—
General Accuracy Over Time
Sentiment Distribution
Source Reliability
Volume Impact
Daily General Sentiment Timeline
Ticker Analysis

Select a Ticker

Choose a ticker from your watchlist
Analyze
Trading Days
—
Ticker Accuracy
—
awaiting selection
Combined Accuracy
—
awaiting selection
General Accuracy
—
awaiting selection
Avg Confidence
—
awaiting selection
Total Signals
—
awaiting selection
Today's Signal Dashboard
Outcome by Agreement Combo
Cumulative Hypothetical Returns —% · $—
Accuracy by Market Regime
Avg Return by Confidence × Direction
Return Magnitude Distribution
Accuracy by Agreement Pattern
Daily Signal Timeline
Portfolio

Watchlist

0 tickers tracked
Analysis

Compare Tickers

Select up to 8 tickers to compare
Compare

Select tickers above to view side-by-side stats

Select tickers above to compare confidence calibration

Select tickers above to compare their accuracy over time

Select tickers above to compare returns by signal direction

Select tickers above to view streak & momentum heatmap

Select tickers above to compare regime accuracy across tickers

Information

About

Financial Sentiment Signal Detector
About This App
A full-stack, serverless financial sentiment analysis system that fine-tunes a LLaMA 3.1 8B model on AWS SageMaker using QLoRA, then deploys it to classify financial news articles in real time. Every trading day, an automated pipeline ingests articles from Marketaux, runs them through the fine-tuned model to determine sentiment, and feeds the results into a three-agent LangGraph pipeline powered by Amazon Bedrock — where each agent independently analyzes sentiment, evaluates market context, and generates a final BUY / SELL / HOLD signal with a confidence score. Signals are then backtested against actual closing prices from Yahoo Finance to measure prediction accuracy over time.
Tech Stack
Built with Node.js, Python, Serverless Framework v4, and a fully serverless AWS architecture including Lambda, DynamoDB, SageMaker, Bedrock, API Gateway, EventBridge, S3, CloudFront, Cognito, and SNS. The multi-agent signal pipeline uses LangChain and LangGraph. The dashboard is built with vanilla JavaScript and Chart.js.
Source Code
This project is open source.
View on GitHub
Model Architecture
Fine-tuned LLaMA 3.1 8B using QLoRA (4-bit NF4 quantization, rank 32) on the Financial PhraseBank dataset — 2,264 labeled sentences mapped to bullish, bearish, and neutral classes. Training runs on ml.g5.xlarge via SageMaker with HuggingFace PyTorch DLC.
LLaMA 3.1 8B QLoRA NF4 4-bit rank 32
Evaluation Pipeline
After each training job completes, an EventBridge rule triggers a SageMaker Processing Job that loads the model artifacts, runs greedy inference on the held-out test split, and writes a results.json to S3. Metrics include exact-match accuracy, macro/per-class F1, and BERTScore (roberta-large, rescaled).
scikit-learn bert-score roberta-large ml.g5.2xlarge
Account

Settings

Manage your account and preferences
?
Account Information
Your account details and registration information.
Name
—
Email
—
Member Since
—
Change Password
Update your password. You'll need your current password to set a new one.
Sign Out
End your current session and return to the login screen.
Delete Account
Permanently delete your account and all associated data. This action cannot be undone.
Appearance
Choose how the app looks. Select a theme or follow your system preference.