Show HN: Pebble Finance – Explain
pebble.finance
Hello HN,
We're back! You might remember us, Pebble Finance, from when we introduced our dynamic investment theme construction tool (https://pebble.finance/themes). Today, we're excited to share with you the next phase of our journey—demystifying investment performance with Pebble Explain.
Introducing Pebble Explain
Pebble Explain is our newest product, developed using advanced summarization, clustering techniques, and powered by OpenAI's LLM. At its core, it seeks to answer a fundamental question: Why did a particular security perform the way it did? By sifting through price movements, financial news, documents, and more, Pebble Explain provides insights on the performance of securities over different timeframes—from a single day to up to six months in the past.
For now, you can look into the explanation of a single security per timeframe. But, if you're eager to get a holistic understanding of your entire portfolio, simply connect your brokerage account via Plaid. We'll serve up insights for every security you hold.
Current Scope and Future Plans
At this stage, our focus is solely on US equities. For those of you keen on insights for ETFs, stay tuned—support for ETFs is on our roadmap!
Challenges We’ve come across
We faced issues primarily related to speed, quality, and token limits. Handling large date ranges required a careful selection process to eliminate unnecessary articles, focusing instead on the ones that best explain the company's stock movements. As we added more articles, not only did processing time increase, but we also edged closer to OpenAI's token constraints. While GPT-4 offers superior quality, its slower response led us to experiment with prompt tuning. This enabled us to achieve results similar to GPT-4, but by leveraging the GPT-3.5-Turbo-Instruct model which is faster and cheaper. We're eager to hear your feedback, thoughts, and suggestions. Dive in, explore Pebble Explain, and let us know what you think!
Thanks for reading!
The Pebble Finance Team