Multi-Agent Financial Analysis Across Data, APIs, and Charts
Financial analysis lived in separate databases, vendor APIs, and charting tools, so answering a question meant stitching those sources together by hand.
The Problem
Analysts already had the raw material — warehouse tables, market and fundamentals APIs, and charting tools — but none of it worked as one system. A single question meant hopping between query consoles, API docs, and chart UIs, then assembling the answer by hand. I helped design and implement an agentic layer that could use those resources together.
The Architecture
flowchart TB
subgraph sources [Analysis Resources]
DBs[(Databases)]
APIs[Market and Fundamentals APIs]
Charts[Charting Resources]
end
subgraph agents [Multi-Agent System]
Planner[Planner Agent]
DataAgent[Data Agent]
MarketAgent[API Agent]
ChartAgent[Charting Agent]
Synthesizer[Synthesis Agent]
end
subgraph output [Analysis Output]
Brief[Analysis Brief]
Figures[Charts and Figures]
end
Planner --> DataAgent
Planner --> MarketAgent
Planner --> ChartAgent
DataAgent --> DBs
MarketAgent --> APIs
ChartAgent --> Charts
DataAgent --> Synthesizer
MarketAgent --> Synthesizer
ChartAgent --> Synthesizer
Synthesizer --> Brief
Synthesizer --> Figures Multi-Agent Tooling
The system treats each source as a capability, not a destination:
- Planner Agent: Breaks an analysis question into retrieval, computation, and visualization steps
- Data Agent: Queries internal databases for positions, history, and reference data
- API Agent: Pulls live and historical series from external financial APIs
- Charting Agent: Renders the relevant views from the charting stack
- Synthesis Agent: Combines those results into a brief an analyst can actually use
Agents share context so a chart request can follow from a database result, and an API pull can fill gaps the warehouse does not have.
Tech Stack
- Multi-agent orchestration — Planner plus specialist agents with tool access
- SQL / warehouse access — Structured financial data
- External APIs — Market and fundamentals feeds
- Charting resources — Figures generated as part of the analysis path
The Impact
The work was design and implementation of the analysis system itself: a multi-agent architecture that can move across databases, APIs, and charting resources in one pass, instead of leaving that stitching to the analyst.