Agentic Sensor Analytics

An agentic AI system for natural-language analytics of smart building sensor data.

Overview

This project enables users to query building sensor data using natural language questions. The system combines a local LLM for intent extraction with deterministic analytics tools to provide accurate, verifiable results.

Architecture

The system uses a five-layer agentic architecture:

1. User Query → Natural language input
2. Intent Interpretation → LLM extracts structured task specification
3. Agent Planning → Validates parameters and creates execution plan
4. Tool Execution → Deterministic analytics compute results
5. Result Explanation → LLM converts results to natural language

This separation ensures reliable computations while maintaining natural language interaction.

Key Features

• Natural Language Interface: Ask questions in plain English
• Deterministic Analytics: All calculations are verifiable and reproducible
• Local LLM: Uses Ollama for privacy-preserving inference
• Real Sensor Data: Connects to SMT Analytics API for Peavy Hall building data
• Full Transparency: Complete execution traces for every query

Quick Start

Prerequisites

• Python 3.10+
• Ollama installed and running
• Access to SMT Analytics API

Setup

# Clone repository
git clone <repository-url>
cd agentic-sensor-analytics

# Install dependencies
pip install -r requirements.txt

# Pull LLM model
ollama pull llama3.1:8b

# Run app
streamlit run .\ui\app.py

OR

# Clone repository
git clone <repository-url>
cd agentic-sensor-analytics

# Docker - make sure Docker is running!
docker compose up --build

Configuration

Create .env file from .env.example with SMT API credentials.

Author

Sean Clayton