eddyzzl/marvis-risk-agent

MARVIS-Agent: all-purpose credit risk agent for model development, validation, data processing, feature engineering, and strategy workflows.

What it solves

MARVIS-Agent is a local-first platform designed to automate the complex, fragmented workflow of credit risk analysis and strategy development. It replaces the manual chain of disconnected scripts, notebooks, and spreadsheets by providing a governed workbench where data processing, model development, and risk strategies are linked with a clear audit trail.

How it works

Users describe their risk goals in natural language. The Agent then clarifies requirements, builds a validated workflow plan, and executes deterministic tools to perform calculations. Unlike a standard chatbot, MARVIS uses the LLM for orchestration and explanation, while the actual metrics (like KS, AUC, and PSI) are calculated by hard-coded platform kernels to ensure mathematical accuracy and auditability. The system includes human-in-the-loop "responsibility gates" for high-impact decisions.

Who it’s for

It is built for risk teams and credit analysts who need to develop governed models and strategies while keeping sensitive data local and maintaining a strict audit trail for regulatory or internal review.

Highlights

  • End-to-End Strategy Workflow: A seven-step process covering everything from project understanding and sample design to impact measurement and final report generation.
  • Governed Model Development: Tools for feature engineering, training multiple recipes (including PMML export), and model validation with structured evidence.
  • Deterministic Execution: Ensures that risk metrics and financial calculations are performed by code, not guessed by an LLM.
  • Local-First Architecture: Keeps files, task states, and outputs in a controlled local workspace to maintain data privacy.
  • Audit-Ready Deliverables: Generates professional reports in Excel, Word, JSON, and Markdown formats with full provenance and lineage.

Related

  • Project
    google/mantis

    Mantis is an open‑source, AI‑driven toolkit that automates security reviews of code. It uses LLM agents to build threat models, generate exploit tests, and suggest patches, all within a sandboxed, configurable workflow. The project is experimental and intended only for isolated environments; all findings require manual verification.

  • Project
    agnt-gg/agnt

    AGNT is a local‑first, self‑hostable desktop app that bundles AI agents, visual workflows, long‑running goals, memory, plugins and a local HTTP API into one “agent operating system”. Built with Electron, Vue, Express and SQLite, it supports 15+ LLM providers, 60+ workflow nodes, plugin templates and SkillForge for self‑improving skills. Ideal for developers or teams that need durable, inspectable AI automation on‑premise.

  • Project
    microsoft/agent-governance-toolkit

    Agent Governance Toolkit is a Microsoft‑maintained open‑source library that sits between autonomous AI agents and the tools they call, enforcing deterministic, auditable policies (allow/deny/require‑approval) written in YAML/OPA. It provides language‑specific SDKs (Python, TypeScript, .NET, Rust, Go), a CLI (`agt`) for compliance checks, and adapters for major agent frameworks, enabling production‑grade governance, attribution, and tamper‑evident logging.

  • Project
    AI4Finance-Foundation/FinRobot

    FinRobot is an open‑source, multi‑agent AI platform for high‑stakes financial analysis. It separates deterministic Python‑based valuation (DCF, LBO, etc.) from LLM‑generated narration, orchestrates specialized agents (data, analysis, modeling, synthesis, reporting, plus debate agents), and ships three versions: a production desktop app (V2), a self‑hosted web report generator (V1), and an educational AutoGen library (V0). The system integrates multiple market data sources, provides 56 analyst playbooks, and can be run locally via a CLI, a browser UI, or a native Tauri desktop client.

  • Project
    lsdefine/GenericAgent

    A minimal, self-evolving autonomous agent framework that allows LLMs to control a local computer and autonomously grow a personal skill tree from a small set of atomic tools.