Atlas AI System Architecture
Our architecture focuses on grounded financial intelligence, processing data deterministically before leveraging LLMs for natural language explanation.
graph TD
CSV[CSV/JSON Import] --> Parser[lib/ingest.ts]
Parser --> Norm[lib/analytics.ts]
Norm --> Dash[Dashboard/KPIs]
Norm --> Chart[Volume Chart]
Norm --> Chat[Ask Atlas API]
Chat --> NVIDIA[NVIDIA LLM API]
NVIDIA --> Streaming[ReadableStream Response]
Streaming --> UI[Chat UI]
Data Pipeline
- Ingestion: Client-side parsing using standardized Zod schemas ensures data integrity before it ever leaves the user's browser.
- Analytics Layer: Deterministic metrics (Net Cashflow, Volatility, Anomalies) computed via
analytics.ts. - Grounded AI: Context from the analytics layer is injected into the prompt, ensuring the LLM responses are constrained to the user's provided data.