Deep Diligence

Inspiration

I loved Perplexity for its timely news feed and strong finance knowledge. I wanted a dashboard where multiple AI-led company conversations and reports can be saved for ongoing research. Building it as a single complex prompt, complete with live Perplexity and Alpaca keys, felt like the ideal way to push a front-end only prototype to the limit.

What it does

• Deep Research streams a full equity report enriched with fundamentals, charts, and structured commentary.
  
• Casual Chat answers quick questions with a preference for very recent news.
  All sessions and reports are cached locally so you can revisit, compare, and update research without leaving the browser.

How we built it

• React, Vite, and Tailwind power a lightweight, component driven UI.
  
• Perplexity’s /chat/completions endpoint supplies AI answers, switching between sonar and sonar-deep-research.
  
• Alpaca data endpoints feed price bars and fundamentals, visualised with Chart.js.
  
• LocalStorage preserves settings, recent queries, and saved reports, allowing us to postpone any backend work.
  
• A strong system prompt enforces a journalistic tone, inline citations, and Markdown structure.

Challenges we ran into

• Parsing and slimming Alpaca JSON to fit within prompt token limits.
  
• Balancing response depth with speed; deep research mode needs more tokens yet must still stream smoothly.
  
• Keeping API keys secure during local testing.
  
• Preventing visual clutter while presenting charts and long form analysis side by side.

Accomplishments that we're proud of

• Built a fully interactive equity research tool entirely from a single prompt.
  
• Achieved seamless hand off between chart generation and AI commentary.
  
• Implemented automatic source citations inside Perplexity responses with no extra code.
  
• Preserved conversation context and saved reports across sessions with zero backend.

What we learned

• Perplexity’s deep research model excels when you preload structured fundamentals and supply a clear analysis rubric.
  
• Careful prompt engineering can replace a surprising amount of glue code in early prototypes.
  
• Even complex finance dashboards can begin as front end only apps, reducing the barrier to iteration.

What's next for Deep Diligence

• Add secure user authentication and cloud storage for cross-device report syncing.
  
• Expand data integrations to include macro indicators and alternative datasets.
  
• Enable comparative views across multiple tickers in a single report.
  
• Offer PDF export and automated email digests of saved research highlights.

Built With

• bolt.new
• perplexity