A voice agent that listens for commands and controls industrial IoT devices — OpenAI Whisper for transcription, LLM command parsing, and MQTT to edge devices for real-time action.
On a factory floor, hands are busy and screens are far away. Operators needed to control equipment by voice, reliably, in a noisy environment with zero tolerance for a misfired command.
The challenge
Industrial voice control has to survive ambient noise, understand loose phrasing, and translate it into exact, safe device actions in real time — a hard bar for off-the-shelf assistants.
Our approach
Whisper transcribes speech even in noisy conditions; an LLM parses intent from natural phrasing into a structured command; and MQTT pushes that command to edge devices for immediate execution. The loop is tight enough to feel instant and structured enough to stay safe.
Under the hood
- Speech — OpenAI Whisper transcription
- Understanding — LLM command parsing to structured actions
- Transport — MQTT to edge devices
- Execution — Real-time action on industrial hardware
What we built
- Noise-robust voice transcription for the shop floor
- LLM intent parsing from natural, loose phrasing
- MQTT command delivery to edge devices
- A real-time speak-to-action control loop
Results
- Operators control equipment hands-free and eyes-up
- Commands execute in under a second, on the edge
A Lightning Leap case study — one of the AI systems we've designed, built, and shipped.