The problem AquaCult solves

The Crisis: "Blind" Farming

India’s 14 million+ aquaculture farmers operate in an information black box. They rely on guesswork to manage invisible water toxicity and react to diseases only after fish start dying. This reactive approach leads to a staggering 40% crop mortality rate, compounded by barriers of literacy, connectivity, and market access.

The Solution: AquaCult

AquaCult is an end-to-end "Blue-Collar Operating System" available on iOS and Android. We replace guesswork with a closed-loop ecosystem that Predicts, Detects, and Resolves crises in real-time.

1) Our "Hero" Technology

• Pocket Vet: Our computer vision model enables farmers to diagnose diseases (e.g., Red Spot, Fin Rot) instantly by snapping a photo, with no lab required.
• IoT Forecasting: We go beyond monitoring. Our Hybrid AI predicts lethal ammonia/pH spikes up to 72 hours in advance, giving farmers time to take action.

2) Radical Accessibility (Designed for the Next Billion)

• Voice-First & Multilingual: Farmers interact using native voice narration in their local dialect, not complex text.
• SMS Lifeline: Critical alerts are delivered via SMS, ensuring safety even in low-network or 2G rural zones.
• Semantic UI: A color-graded interface (Red/Green indicators) reduces cognitive load by shifting from reading data to recognizing patterns.

3) The Economic Loop

• Hyper-Local Marketplace: We connect diagnosis to cure by enabling AI-recommended treatments from local sellers through an integrated portal.
• AquaCult Certified Farms: We aggregate successful farmers under the "AquaCult Certified" banner, building a trust layer that helps them sell harvests at premium rates to export markets.

Challenges we ran into

Challenges We Ran Into

1) The "Heavyweight" Integration: ResNet50 on the Edge

• Problem: ResNet50 was accurate but too heavy, causing crashes on lower-end devices.
• Struggle: Uploading high-res images to our FastAPI backend timed out on slow rural-like networks.
• Fix: Client-side image compression + asynchronous backend inference queue, reducing click-to-diagnosis from 12s to under 3s.

2) The "Tower of Babel": True Multilingual Support

• Problem: Direct translations broke critical aquaculture terminology and confused users.
• Fix: Switched to Gemini 1.5 Flash with a prompt to translate using rural-friendly local analogies, then routed output into TTS for accurate voice delivery.

3) Voice-First Latency

• Problem: STT → LLM → TTS caused ~10s response lag and poor UX.
• Fix: Optimistic UI updates + concise prompts, cutting round-trip latency by 40%.

4) Data Scarcity for Rare Diseases

• Problem: Limited datasets for Indian fish diseases (e.g., Argulosis).
• Fix: Data augmentation (rotate, flip, noise) expanded 50 images to 500 training samples, reducing overfitting.