AI agent platform for KPI analytics

The problem

A mid-size organization can track 8,000+ key performance indicators (KPIs) across CRM, accounting and ERP systems. Managers want to ask questions in plain language and get answers that come from the real data, not from a guess.

The orchestrator

The core is an LLM tool-use orchestrator for multi-step workflows. Its tools cover KPI queries, SQL, research, memory and document generation.

• Parallel tool execution and streaming. The user sees partial results while slower tools run.
• Multi-provider routing. One interface routes to Anthropic Claude, OpenAI, Gemini or self-hosted Llama.
• Compact context. A new KPI encoding halved the per-KPI prompt payload, from 80 to 40 tokens.

The multi-agent system

A second system runs on Google Agent Development Kit (ADK). A master agent extracts intents in parallel. Then it coordinates 7+ specialized agents for search, research, SQL generation, data retrieval, reporting and auditing. These agents work over SugarCRM, QuickBooks, NetSuite and HubSpot.

Tools over MCP

• KPI data tool. One core with three interfaces (CLI, Python API and an MCP server). The server exposes 16+ Model Context Protocol tools for navigation, data retrieval, time series, discovery and data-quality checks.
• Forecasting platform. 13 statistical, machine learning and foundation models (TimeGPT, Chronos-2, TimesFM) as 40+ MCP tools. It adds walk-forward backtesting, four uncertainty quantification methods, ensembling, hierarchical reconciliation and scenario analysis.

The knowledge graph

A multi-tenant, LLM-augmented knowledge graph unifies KPIs, goals, data lineage and decisions for downstream AI pipelines. Dual-model adversarial judging, what-if simulation and human-in-the-loop questions raised audited KPI classification accuracy from 21% to 79% in staging.