AQUA-1B: Aquaculture-Specific Small Language Model

AQUA-1B, developed by Kurma AI, is a 1-billion parameter Small Language Model (SLM) built upon Google DeepMind's Gemma 3 1B base. It is the first lightweight model specifically tailored for the aquaculture domain, focusing on real-time operations, IoT sensor data processing, and autonomous system control.

Key Capabilities

Edge-Ready Intelligence: Optimized for low-power, real-time inference on embedded devices like Raspberry Pi and Jetson Nano.
Agentic Task Execution: Supports multi-step, agent-based workflows for sensor checks, feeding triggers, and autonomous health checks.
IoT-Aware Reasoning: Natively understands and reasons over sensor data inputs (e.g., temperature, pH, TDS) for rapid decision-making.
Robotic Automation Control: Designed to interact with robotic systems, including underwater and mobile pond inspectors.
Autonomous Alerting Systems: Powers local alert mechanisms (SMS, Telegram bots, MQTT) for critical parameter thresholds.
Field-Deployable Decision Engine: Enables autonomous operation in remote hatcheries and ponds, even in offline conditions.

Training and Data

AQUA-1B was fine-tuned using a LoRA-based Supervised Fine-Tuning (SFT) approach on approximately 3 million real and synthetic Q&A pairs, totaling around 1 billion tokens of high-quality, domain-specific data. This dataset includes extension worker–farmer dialogues, FAO/ICAR/NOAA research, and species-specific culture methods.

Good for

On-device decision-making in aquaculture operations.
Real-time alert generation based on water quality parameters.
Automating tasks like feeding routines and water exchange scheduling.
Controlling robotic systems for inspection and monitoring in ponds and RAS.
Deployments in low-connectivity or offline environments requiring autonomous operation.