QuantNova

Inspired by the need to democratize institutional-grade financial research, QuantNova is an autonomous "Sovereign Analyst" built on Amazon Bedrock that transforms how investors interact with market data. By leveraging a multi-agent architecture, the system coordinates specialized Amazon Nova models to perform complex tasks: a Researcher ingest thousands of pages of SEC filings using Nova’s 1M token context, a Chartist employs multimodal vision to decode technical price patterns, and an Auditor scans for bearish risks to mitigate bias. We utilized Nova Act to move beyond brittle HTML scraping, allowing the agent to navigate live web dashboards with human-like intent, while Amazon OpenSearch ensures every insight is grounded in RAG-verified citations. Throughout development, we overcame "reasoning loops" by fine-tuning Nova 2’s Thinking Intensity and successfully stress-tested the model's massive context window to find hidden risks in multi-year 10-K reports. This journey taught us that agentic reliability hinges on "Extended Thinking"—allowing the AI to deliberate before acting—resulting in a trusted system that adheres to Bedrock Guardrails. Looking ahead, we plan to integrate Nova Sonic for real-time voice briefings and Nova Reel to generate automated, 60-second video market summaries, making QuantNova the ultimate omni-modal investment companion.

Built With

• agents
• python
• streamlit