FinanceSight

FinanceSight generate deep narrative synthesis from financial data and model predictive scenarios for superior investment decision support.

Comment

FinanceSight Project Documentation

Inspiration

FinanceSight utilizes the Sonar API from Perplexity in order to obtain information. It employs deep sonar-pro and rapidly collects news, risks, and opinions. It uses sonar-deep-research for our 'Stock Research Engine,' which forescasts storck trajectories in the future time period. It monitors shifting narratives, discovers possible triggers and constructs future narratives out of previous examples and the latest cues from the web. The full citation functionality of Perplexity plays a very important part in all our analyses and delivers necessary clarity and trustworthiness for each financial insight provided to the client.

What it does

FinanceSight is an AI powered financial intelligence platform designed to help users cut through market noise and gain a deeper, more forward looking understanding of investments.

At its core, FinanceSight utilizes Perplexity's Sonar API to:

  • Synthesize News & Events: Provides concise summaries of relevant developments for user specified companies or themes (Feature 1).
  • Analyze Market Sentiment: Offers insights into the prevailing sentiment and its drivers for specific assets (Feature 2).
  • Identify Key Risks: Summarizes key risk factors associated with companies or sectors (Feature 3).
  • Finance Chatbot: An expert in finance and stock that can answer any question (Feature 4).
  • Stock Research.: Delivers comprehensive overviews of a stock, create narratives across web sources and models it future projections.

Architecture

Our flagship innovation is the "Stock Research" This goes beyond simple analysis by:

  • Tracking how investment narratives evolve over time across a multitude of web sources.
  • Identifying emerging signals and historical analogues.
  • Modeling plausible future narrative pathways for companies or themes, highlighting potential triggers and implications.

Stock Research

Essentially, FinanceSight aims to help users not just understand the current financial story, but to anticipate its next chapters.

How we built it

FinanceSight is built by harnessing the unique capabilities of Perplexity's Sonar API as its primary intelligence layer.

Core Technology:

Perplexity Sonar API: We utilize various Sonar models. For instance, sonar-pro is used for its balance of comprehensive search and synthesis for features like news summarization and sentiment analysis. Our "Stock Research Engine" leverages sonar-deep-research for comprehensive data gathering, historical context, narrative shifts, and constructing plausible future pathways.

Frontend: The user interface is developed using React with the shadcn/ui component library, creating a clean, responsive, and intuitive experience. We designed a multi column layout with a dedicated right hand panel for specialized tools (like the tabbed feature toolkit for news, sentiment, and risk, and a separate section for thematic exploration).

Sonar API Integration

For the core information synthesis in FinanceSight, I primarily utilized sonar-pro.

  • For News & Events Synthesis (Feature 1), I tasked sonar-pro with conducting real-time web searches across financial news outlets, company press releases, and relevant publications. It then synthesized these findings into concise summaries of key developments for specified companies or themes.

  • In Market Sentiment Analysis (Feature 2), I used sonar-pro to scan a variety of online sources to identify textual indicators of market sentiment. It was effective in summarizing the prevailing mood and extracting the key underlying drivers for that sentiment.

  • To Identify Key Risks (Feature 3), I leveraged sonar-pro's ability to search through diverse materials like summaries of official filings and in-depth news analyses. It synthesized mentions of potential risks into a coherent list pertinent to the queried company or sector.

  • For the Finance Chatbot (Feature 4), sonar-pro served as the backbone. It enabled the chatbot to provide comprehensive and up-to-date answers to a wide spectrum of financial queries by performing real-time web searches and synthesizing information into clear responses.

For the more advanced Stock Research (Feature 5), which involved creating comprehensive overviews, constructing evolving narratives, and modeling potential future projections, I employed sonar-deep-research.

  • Its reasoning capabilities were crucial for building comprehensive overviews and narratives. I directed it to gather information from a multitude of disparate web sources, and it used its Chain-of-Thought process to connect these pieces and construct cohesive, evolving stories around a stock or company.

  • When modeling future projections for our qualitative narrative foresight, I utilized sonar-deep-research's advanced analytical abilities. It helped identify historical analogues and emerging signals, reasoning through these inputs to construct plausible future narrative pathways and outline potential triggers or implications for each scenario.

Challenges we ran into

  • Synthesizing Coherent Narratives: Distilling vast, often conflicting, information from diverse web sources into concise and truly insightful narratives is a significant AI challenge. Ensuring the AI captures nuance and avoids oversimplification was key.

  • Modeling Narrative Evolution: Developing robust logic for the "Stock Research Engine" to accurately identify genuine shifts and plausible future pathways, rather than just highlighting coincidental events, proved complex. Distinguishing signal from noise in narrative trends was an iterative process.

  • Prompt Engineering for Specificity: Crafting precise prompts for the Perplexity Sonar API to extract very specific information (e.g., a 2-sentence summary for the last 30 days, or causal inferences for narrative shifts) required significant experimentation.

  • UI for Complex Data: Designing an intuitive UI (with shadcn/ui and React) to present complex, multi-faceted analysis like evolving narratives and potential future scenarios in a way that is easily digestible for users was a considerable design challenge.

Accomplishments that we're proud of

  • Successfully Leveraging Perplexity Sonar API for Advanced Narrative Insight: We're proud of demonstrating how the Sonar API can be used for more than just Q&A or basic summarization, pushing it towards sophisticated narrative analysis and foresight generation.

  • Conceptualizing the "Stock Research Engine": Developing a genuinely innovative feature concept that aims to provide predictive insights into how investment stories might unfold, a significant step beyond current commercial offerings.

  • User Centric UI Design: Creating a clean and intuitive interface using React and shadcn/ui that can make complex financial information more accessible and actionable for users.

  • Proof of Concept for Information Overload Solution: Demonstrating a tangible way to help investors combat information overload by providing synthesized, context rich, and forward looking intelligence.

What we learned

  • The Power of Perplexity Sonar API: We gained deep appreciation for the Sonar API's capabilities in real time web search, deep research, and especially its reasoning and synthesis power for complex financial topics.

  • Nuances of Financial Narrative Analysis: Understanding that financial narratives are multi layered and influenced by a vast array of factors, requiring sophisticated AI to unravel.

  • Importance of Prompt Engineering: The quality and specificity of prompts are paramount in guiding the AI to produce the desired analytical depth and concise outputs.

  • Iterative Development in AI Projects: Realized that building AI powered features involves continuous iteration on prompts, data processing logic, and result interpretation.

  • Bridging AI Capabilities with User Needs: The challenge and importance of translating powerful AI outputs into user friendly interfaces and genuinely valuable insights for the end user.

What's next for FinanceSight

  • Develop & Refine the Stock Research Engine: Fully implement and rigorously test this core innovative feature, focusing on the accuracy of pathway modeling and the clarity of insights.

  • Expand Behavioral Finance Integration: More deeply integrate principles of behavioral finance to help users identify and mitigate cognitive biases based on narrative framings.

  • Enhance Personalization: Allow users to create highly customized dashboards, define specific narrative triggers for alerts, and tailor the foresight parameters to their unique investment strategies and risk profiles.

Built With

Share this project:

Updates