% ============================================================================ \section{Introduction} \label{sec:introduction} % ============================================================================ The rapid maturation of large language models (LLMs) has transformed expectations about what conversational software can accomplish in finance. Models such as the Claude, GPT, and Gemini families now produce fluent commentary on earnings, valuations, and macroeconomic conditions, and a growing literature documents their ability to extract signal from financial text \citep{lopezlira2023chatgpt,kim2024financial,wu2023bloomberggpt}. Yet fluency is not the same as rigor. A model that \emph{describes} a value-at-risk calculation in prose has not \emph{computed} one; a model that \emph{recalls} a stale price from pre-training has not \emph{retrieved} the current quote; and a model that narrates a Monte Carlo experiment has not run the simulation whose numbers a portfolio manager would act upon. The central engineering problem in applied financial AI is therefore not generation but \emph{grounding}: how to bind a probabilistic text generator to live data, deterministic computation, and reproducible artifacts without sacrificing the conversational interface that makes the technology accessible. This paper documents the methodology and innovations of \textbf{VQuant}, a production-grade web platform built to address precisely this problem. VQuant accepts a natural-language request---``analyze NVDA's fundamentals, technicals, and run a Monte Carlo simulation,'' for example---and returns a complete, multi-section quantitative report containing live data, computed figures, and downloadable documents. The platform is implemented as a full-stack TypeScript monorepo: a React front end, a modular Express back end, and an isolated Python scientific-computing layer. At its core, the system orchestrates a reasoning-augmented LLM through an \emph{agentic loop} that exposes more than two hundred financial and analytical tools to the model, executes the model's tool calls deterministically, streams the results back in real time, and synthesizes a final report that can be exported in five document formats. The motivation for studying this system is fourfold. \emph{First}, VQuant operationalizes a design philosophy that we believe is the correct one for high-stakes numerical domains: the LLM is a \emph{planner and synthesizer}, never a calculator. Every number in a VQuant report originates from a deterministic source---a market-data API, a closed-form pricing formula, or auditable Python code---so that the analysis is reproducible and the model's well-known propensity for numerical hallucination is structurally contained. \emph{Second}, the platform confronts the practical constraints of agentic systems---bounded context windows, tool-call latency, rate limits, and cost---with concrete algorithmic solutions whose logic generalizes beyond finance. \emph{Third}, VQuant integrates an unusually broad tool ecosystem (over 265 market-data endpoints, four web-research services, and a seventeen-library Python sandbox) under a single function-calling protocol, providing a useful case study in tool taxonomy and prompt-level routing. \emph{Fourth}, the system closes the loop from query to deliverable by generating publication-ready PDF, Word, \LaTeX{} Beamer, and spreadsheet outputs from a single internal representation, illustrating a reusable pattern for automated report generation. Our contributions are methodological and architectural rather than empirical. We make four specific contributions. \textbf{(1)~Agentic orchestration with constrained tool use.} We describe a streaming tool-calling loop (Section~\ref{sec:methodology}) in which a strategically engineered system prompt routes requests to an appropriate subset of more than two hundred tools, batches tool execution to respect external rate limits, and surfaces the model's intermediate reasoning to the user in real time. \textbf{(2)~Context management under a token budget.} We formalize an adaptive-summarization algorithm that preserves coherence across effectively unbounded conversation histories by compressing older turns once a token threshold is crossed, while retaining the most recent exchanges verbatim. \textbf{(3)~A single-call computation paradigm.} We argue that consolidating data acquisition, numerical estimation, and visualization into one sandboxed Python execution---rather than the sequential fetch-then-analyze pattern typical of naive agents---materially reduces end-to-end latency, and we describe the sandbox and its safety properties (Section~\ref{sec:capabilities}). \textbf{(4)~A unified multi-format document pipeline.} We detail how a single Markdown artifact is transformed into PDF (headless-browser rendering with mathematical typesetting), DOCX (document converter), Beamer slides (model-structured \LaTeX{} compiled to PDF), and spreadsheets with live formulas. Our emphasis throughout is on \emph{mechanism}. We describe each component as it is actually built---the byte-level streaming protocol, the recursion that drives the loop, the throttle that batches tool calls, the two-stage machinery that keeps the context within budget, the correction that keeps token accounting honest, and the inter-process boundary that executes model-authored code---so that the design is reproducible rather than merely sketched. The classical quantitative methods the platform automates (option pricing, volatility modeling, simulation, and risk metrics) are standard and not the contribution of this paper; we record them compactly in an appendix and otherwise keep the focus on the architecture. The remainder of the paper is organized as follows. Section~\ref{sec:literature} situates VQuant within the literatures on LLM agents, tool use, retrieval augmentation, and AI in finance. Section~\ref{sec:architecture} presents the system architecture, data layer, and external integrations. Section~\ref{sec:methodology} develops the agentic orchestration methodology, including the tool taxonomy, context-management algorithm, and single-call execution paradigm. Section~\ref{sec:capabilities} formalizes the embedded quantitative methods and the document-generation pipeline. Section~\ref{sec:evaluation} discusses performance, reliability, security, and limitations. Section~\ref{sec:conclusion} concludes and outlines directions for evaluation and extension.