Deep Finance Research

Deep Finance Research is an AI-powered equity research application that turns a company name or ticker into an investor-style report with supporting charts and exportable deliverables.

The app combines a Deep Agents orchestration layer, market data retrieval, web research, and document generation to produce structured reports that look closer to analyst work than a generic chat response.

Inspiration

Retail investors and small teams often have access to fragmented market data, scattered news, and shallow AI summaries, but not to a disciplined research workflow. Market Insight was built to close that gap by giving users an AI system that behaves more like a research desk: structured, source-aware, and capable of producing outputs that are useful beyond a chat window.

What it does

Market Insight turns a company name or ticker into an investor-style research package.

• It analyzes company fundamentals, business context, industry structure, competition, valuation, and risks.
• It uses specialist agents instead of a single generic assistant.
• It produces structured markdown reports, charts, PDFs, and presentations.
• It exposes the workflow through both a Streamlit interface and a FastAPI backend.

How we built it

• We used Deep Agents to create an orchestrator plus specialist sub-agents for company data, industry research, valuation and risks, and presentation generation.
• We connected Yahoo Finance-based tools for financial data and EXA for broader web research.
• We added export utilities so the final research can be turned into PDF and PowerPoint artifacts.
• We exposed the same underlying agent graph through a Streamlit UI for interactive use and a FastAPI API for programmatic access.

Challenges we ran into

• Keeping long-form research structured enough to feel like an analyst report instead of a loose AI summary.
• Coordinating multiple sub-agents so their outputs complement each other rather than repeat the same findings.
• Balancing live market data with broader web context without making the workflow noisy.
• Turning research output into clean downstream artifacts such as charts, PDFs, and presentations.

Accomplishments that we're proud of

• A working multi-agent finance research workflow with clear role separation.
• A user-facing app that can go from prompt to report to downloadable artifacts.
• A backend design where the same agent system powers both UI and API entry points.
• A product direction that feels more like a research tool than a generic chatbot wrapper.

What we learned

• Multi-agent decomposition is valuable when each role has a clear research boundary.
• Good prompts are not enough on their own; tools, output structure, and orchestration matter just as much.
• Finance use cases benefit from having both structured market data and flexible web research.
• Exportability matters: users often need reports and decks, not only answers in chat.

What's next for Market Insight

• Add authentication and user-specific workspaces so reports, artifacts, and sessions are tied to real users.
• Improve source attribution and confidence signals across every section of the report.
• Expand coverage to portfolio analysis, watchlists, screening workflows, and recurring research updates.
• Add stronger evaluation, testing, and guardrails for financial-report quality and consistency.
• Introduce richer frontend experiences beyond the current Streamlit interface for production deployment.

What the app does

• Accepts a research prompt such as Research Microsoft or Full company report on NVIDIA.
• Breaks the work across specialized sub-agents for company data, industry analysis, valuation, risks, and presentation generation.
• Pulls structured financial data from Yahoo Finance tools and optional web context from EXA.
• Synthesizes findings into a multi-section markdown report.
• Generates charts and can export the final output as PDF and PowerPoint.

Landing page description

The Streamlit landing page presents the app as a research workspace for producing investor-ready reports. It explains:

• what kind of output the user should expect,
• which data sources power the workflow,
• which specialist agents participate in the analysis,
• and that the app can generate charts plus downloadable report artifacts.

Built With

• airbyte
• fastapi