Chat & AI Assistant

The Chat section (/chat) is a conversational research assistant built on a Retrieval-Augmented Generation (RAG) pipeline. Every answer is grounded in the Worldview data corpus — news articles, SEC filings, knowledge graph facts, and your portfolio data.

What you can ask

The AI assistant is designed for financial research questions:

• Instrument-specific — "What's driving NVDA's recent move?"
• Portfolio-scoped — "What risk factors are affecting my top 5 holdings?"
• Macro/sector — "How does a Fed rate cut typically affect bank net interest margins?"
• Comparative — "Compare AAPL and MSFT on free cash flow yield"
• News synthesis — "Summarize the last 3 days of news about Tesla"
• Knowledge graph — "What companies supply chips to Apple?"

How it works

1. Intent classification — the question is classified into a query type (instrument, portfolio, macro, comparative, etc.)
2. Retrieval — relevant context is retrieved from the vector store (articles, filings), knowledge graph, and live market data
3. Generation — DeepSeek R1 Distill 32B synthesizes the retrieved context into an answer
4. Citation — every factual claim is tagged with the source document reference

Model

The chat assistant uses DeepSeek R1 Distill 32B via DeepInfra. This is a 32-billion parameter reasoning model distilled from DeepSeek-R1. It shows its reasoning process before giving the final answer, which helps you evaluate the logic behind responses.

Conversation threading

Each conversation is a thread. Threads persist in your account history. You can:

• Start a new thread (click + New chat or press ⌘N)
• Return to a previous thread from the sidebar
• Rename a thread by clicking its title
• Delete a thread from the ellipsis menu

Related pages

• Asking questions — tips for better prompts
• Citations — how source attribution works
• Slash commands — /ticker, /brief, /compare