-
-
main page
-
msft page
Inspiration
⭐️Historically, the stock market moves on news and emotion faster than anything else. But while institutional investors have access to expensive tools and proprietary feeds to capitalize on these shifts, everyday investors are often left behind — forced to scroll through Reddit, news headlines, and Twitter threads just to keep up.
⭐️That’s why we created Chatalytics — to quantify those emotional market movements in real time and make that insight available to everyone.
⭐️Every day, millions of users and journalists publish opinions about companies online. Our platform captures these moments the second they happen, analyzes the content and sentiment, identifies the companies affected — and predicts the resulting stock price movement, instantly.
⭐️By automating this entire pipeline with advanced machine learning, we’re not just making faster predictions — we’re democratizing access to real-time financial intelligence. Whether you're a casual trader or a data-driven investor, Chatalytics gives you institutional-grade insight, without the institutional price tag.
What it does
⭐️Chatalytics predicts stock price changes in real time by analyzing online news sentiment, social media conversations, and historical stock price patterns.
⭐️Real-time Monitoring: Automatically captures financial news, Reddit posts, and user commentary the moment they're published.
⭐️Sentiment Analysis: Uses advanced NLP models fine-tuned for financial language to evaluate sentiment and relevance.
⭐️Company Mapping: Links posts to the correct stock ticker, even when users refer to companies by nicknames or symbols.
⭐️Price Prediction Engine: Combines sentiment signals with historical price data to forecast short-term stock price movement.
⭐️Fully Automated Pipeline: From data collection to prediction, the entire process runs without manual input, updating as fast as the internet does.
How we built it
- Used machine learning algorithm Gradient Boosting Regression Trees to predict the price change of the stock
- AWS RDS with PostgreSQL for data storage
- Typescript, React and Vite for front end
- Python and Flask for back end
- Fully hosted on AWS EC2 (CHECK OUT THE TRY OUT LINK)
Challenges we ran into
⭐️ Noise in Online Conversations
⭐️Not every mention of a company is meaningful. We had to develop robust NLP filters and relevance scoring models to distinguish between impactful comments (e.g. an earnings leak) and casual chatter (“I like Apple products”).
⭐️Entity Recognition Across Slang and Abbreviations Users often refer to companies in non-standard ways (e.g. “GOOG” instead of “Alphabet” or “$AAPL” instead of “Apple”). Building a custom entity recognition system that could understand tickers, nicknames, and contextual references was a key hurdle.
⭐️Latency in Data Processing To be genuinely “real-time,” we had to optimize our data pipeline to process thousands of new comments per second with minimal lag — which meant engineering around rate limits, streamlining our ingestion system, and scaling our backend.
⭐️Sentiment Ambiguity Sarcasm, irony, and financial jargon make sentiment analysis especially difficult in this domain. We couldn’t rely on off-the-shelf models. Instead, we fine-tuned transformer-based models (like BERT and FinBERT) specifically for finance-related sentiment.
Accomplishments that we're proud of
⭐️Real-Time Sentiment Engine Built a fully automated NLP pipeline that captures and analyzes financial news and Reddit posts in under 3 seconds from publication to prediction.
⭐️High Signal Precision Achieved a 72% accuracy rate in predicting the direction of stock price movement within 30 minutes of high-impact news events.
⭐️Custom Financial Sentiment Model Fine-tuned a transformer-based model (based on FinBERT) specifically for financial language, outperforming baseline sentiment models by 18%.
⭐️Scalable Architecture Deployed a production-ready backend capable of processing over 250,000 text data points per day with minimal latency using AWS and Kafka.
⭐️Real-World Backtesting Ran backtests on over 1M news-comment-price pairs from the past 2 years, demonstrating consistent performance and predictive value.
What we learned
⭐️Sentiment Alone Isn’t Enough Raw sentiment scores can be misleading. Context — like the credibility of the source or the stock’s recent volatility — is crucial for accurate predictions.
⭐️Speed Matters More Than We Thought In financial markets, timing is everything. Even a 5-second delay in processing news can make the difference between a winning and a losing signal.
⭐️Not All Mentions Are Equal A single comment from a top Reddit investor can drive more market movement than 100 random tweets. Weighting sources by influence dramatically improved our model’s performance.
⭐️Financial Language is Nuanced We learned that off-the-shelf sentiment tools don’t understand finance well. Terms like “beat,” “short,” or “burn” have different meanings in this context — which is why we had to train domain-specific models.
What's next for Chatalytics
⭐️Our vision is to become the go-to intelligence layer for real-time market sentiment — empowering traders, analysts, and institutions to make smarter decisions, instantly.
⭐️We believe that in the age of information overload, the edge doesn’t come from having more data — it comes from making sense of it faster and better than anyone else. Chatalytics will be the platform that transforms raw noise into actionable insights, as they happen.
⭐️Expand Source Coverage We're adding more platforms beyond Reddit and major financial news — including X (formerly Twitter), Seeking Alpha, and YouTube transcripts.
⭐️Launch Portfolio Monitoring Instead of searching one ticker at a time, users will soon be able to track sentiment shifts across their entire portfolio, with instant alerts.
⭐️Institutional API We're building an enterprise-grade API to serve hedge funds, quant teams, and trading platforms who want to integrate our real-time sentiment and prediction engine directly into their workflows.
⭐️Mobile App Rollout A mobile-first version of Chatalytics is in development — giving retail traders lightning-fast market insights right in their pocket.
⭐️Sentiment + Technical Fusion Model We're currently testing a hybrid model that combines our sentiment predictions with technical indicators to improve precision and confidence scoring.
Log in or sign up for Devpost to join the conversation.