Finance App

From Pipeline to Product

Modules 1-4 built the engine — ingestion, encoding, retrieval, and a gateway with financial guardrails. But an engine without a dashboard is useless to analysts. This module builds the interface they'll actually use every day.

Financial dashboards have different requirements than general chat UIs. Analysts need data density — numbers, tables, trend indicators — alongside natural language analysis. They need to see the sources behind every claim. And they need to trust the numbers.

The Financial Analysis Agent

The core of the application is a LangGraph analysis agent — an orchestrated workflow that goes far beyond simple question-answering.

When an analyst asks "Compare Q3 margins across our top 5 competitors," the agent doesn't just search and summarize. It executes a multi-step workflow:

Step 1: Parse the Query

Extract structured intent from the natural language query:

Step 2: Identify Peer Group

If the query says "top 5 competitors" without naming them, the agent needs to identify the peer group. This might come from:

Step 3: Parallel Data Fetching

LangGraph's fan-out pattern retrieves data from all sources simultaneously:

This parallel execution is critical for performance. Fetching 5 companies x 4 sources sequentially would take seconds. In parallel, it completes in the time of the slowest single fetch.

Step 4: Calculate Derived Metrics

Raw filing data gives you revenue and cost of revenue. The agent calculates:

These calculations happen on extracted, verified numbers — not by asking the LLM to do arithmetic (which it does poorly).

Step 5: Synthesize Analysis

Feed the structured data through the AI Gateway (Module 4) with a focused prompt:

Step 6: Format Output

Structure the response for the dashboard:

Dashboard UI Design

Data-Dense Layout

Financial dashboards prioritize information density over whitespace. The layout includes:

Streaming with Financial Tables

Standard streaming shows text appearing word by word. Financial streaming is more complex:

Source Citations

Every factual claim in the analysis must link back to its source:

Conversation History

Financial analysts often refine their analysis through follow-up questions:

The chat maintains context, so follow-up questions build on previous results without re-retrieving everything.

Suggested Follow-Ups

After each analysis, the system suggests relevant next questions based on what data was retrieved but not yet discussed:

These suggestions accelerate the analyst workflow — instead of thinking about what to ask next, the system surfaces the most relevant follow-up paths.

What You'll Build