AI-BI Prototype

A data-agent workflow combining semantic routing, guarded code execution, forecasting, and automatic retry.

Personal / Data Agents

A data-agent pattern combining semantic routing, guarded code execution, forecasting, and automatic retry.

Overview

The prototype explores controlled agent execution for business analytics and forecasting workflows.

By separating semantic routing, code generation, and sandboxed execution, the system ensures analytical code is evaluated and retried before rendering final charts or predictions.

The problem

Generated analysis code needs semantic routing, guarded execution, and retry behavior before results can be presented.

My contribution

Combined an intent router, sandboxed code-generation agent, automatic retry, and multiple forecasting models into a cohesive analytics workflow.

System workflow

  1. 01Semantic Routing

    Classify query intent and route to specialized data or forecasting agents.

  2. 02Code Generation

    Synthesize executable analytical code tailored to the target dataset.

  3. 03Guarded Execution

    Execute generated code inside a sandboxed runtime environment.

  4. 04Automatic Retry

    Capture stack traces and auto-correct failed execution paths.

  5. 05Ensemble Forecasting

    Combine Prophet, SARIMA, and XGBoost models with dynamic weighting.

Technical highlights

  1. 01Router Agent with semantic intent classification
  2. 02Sandboxed code-generation data agent with automatic retry
  3. 03Multi-model ensemble with dynamic weight optimization

Links

Next project10ds-monPersonal / macOS Utility

Project image

Interactive demo