Stoqs

What it does

Stoqs is a financial insights and research platform with a unique generative experience. The application offers a beautiful UI and a streamlined UX, allowing users to conduct financial research in multiple threads or use specific companies or news as context for a conversation. We use LLMs as agents to collect, collate and summarize high quality data beyond general search or crawling. Agents decompose queries and plan retrieval from trusted sources - whether it's analyses, raw stats or visualizations. Using Retrieval Augmented Generation indexed SEC Filings, news and earning call transcripts, you get almost insider-level information for your queries.
We adopted an approach to query decomposition specifically for the financial domain. Financial queries typically exhibit interdependencies that require careful planning and resolution. Our decomposition strategy involves a central orchestrator called the Answer & Visual Planner which focuses on breaking down queries into dependent and independent sub-queries, but also on preemptively addressing complex visual requirements. The system iteratively resolves downstream dependencies, processing each sub-query until all dependencies are fully resolved. Upon completion, the orchestrator aggregates the results, collating all textual and visual elements. Our response module synthesizes the information into a coherent output.
Gemini powers our product by allowing us to work with a wide variety of data streams and large context windows.