Edge AI Tutorial with LLM

Описание и применение

Создайте агент на Raspberry Pi, который наблюдает за папкой с телеметрией, отправляет данные в Oxlo.ai, получает структурированные JSON‑отчёты об аномалиях и архивирует файлы.

Задача

Ops teams on lightweight gateways need reasoning on IoT telemetry without managing GPU infrastructure and without cost inflation from token-based billing.

Результат

The agent processes telemetry files every 5 seconds, outputs structured JSON alerts, and archives processed files, enabling lightweight edge reasoning without GPU costs.

Рабочий процесс

• Create project scaffold with agent.py and directories
• Generate mock telemetry files with mock_sensor.py
• Watch the folder and build plain-text prompts
• Define a system prompt that enforces a strict JSON schema
• Wire the Oxlo.ai inference call using llama-3.3-70b
• Run a continuous loop that processes files and archives them
• Start mock sensor and agent in separate terminals
• Observe real-time anomaly reports
• Plan next steps: replace polling with MQTT/inotify, add function calling, batch requests, or upgrade plan

Доказательства

We are building an edge telemetry agent that watches a local folder for IoT sensor dumps, formats the payloads, and calls an LLM via Oxlo.ai to flag anomalies and suggest fixes.

Источник

The five-second sleep keeps CPU usage negligible on a Raspberry Pi.

Источник