Scientific/Finance Tools Module

[a-kore]

PR Type

Feature

Short Description

Adds comprehensive scientific computing toolkit enabling LLM agents to execute Python code for mathematical, scientific, and financial computations. Includes safe code execution engine, support for 12+ libraries (NumPy, SciPy, SymPy, financial packages), tool-assisted scientist agent, and type system improvements. Operates in parametric mode by default (leveraging LLM's built-in knowledge) for minimal latency and token usage.

• ScientificToolKit: High-level orchestration with parametric mode (default), optional RAG/tool-selection (WIP)
• PythonExecutor: Safe code execution with import restrictions and error handling
• ToolAssistedScientist: Enhanced agent with two-stage architecture (code generation → answer formatting)
• Library Support: NumPy, SciPy, SymPy, statsmodels, numpy_financial, py_vollib, PyPortfolioOpt, empyrical, arch
• JSON/LaTeX Handling: Inclusion of json parsing from LLM responses with LaTeX expressions from Ali M.

Tests Added

• 27 unit tests (test_toolkit.py: 18 tests, test_executor.py: 9 tests)
• 8 edge case tests for malformed LLM responses (trailing commas, markdown wrapping, etc.)
• 2 integration tests with real Gemini API (skipped by default to avoid costs)

[saidul-islam98]

Looks good