// ML Trinity — three pillars with motion SVGs const PillarShrink = ({ accent }) => { const t = window.ATTO_TOKENS; const c = accent === 'blue' ? t.blue : t.gold; return ( ); }; const PillarMemory = ({ accent }) => { const t = window.ATTO_TOKENS; const c = accent === 'blue' ? t.blue : t.gold; return ( ); }; const PillarReliable = ({ accent }) => { const t = window.ATTO_TOKENS; const c = accent === 'blue' ? t.blue : t.gold; return ( ); }; const Trinity = ({ accent }) => { const t = window.ATTO_TOKENS; const pillars = [ { num: 'I', name: 'Edge-Cheap', kicker: 'Small Language Models', body: 'Under 4B parameters. 4-bit quantization. Production-grade inference on the NPU in your pocket. Near-zero API cost, 100% private by default.', comp: PillarShrink }, { num: 'II', name: 'Memory-Smart', kicker: 'AI Harness', body: 'Standard LLMs are stateless. Our AI Harness curates and compresses context across sessions — shrinking windows by 60–90%, making every cloud fallback maximally efficient.', comp: PillarMemory }, { num: 'III', name: 'Reliable', kicker: 'HD-NDE Hallucination Guard', body: "Hallucination Detection via Neural Differential Equations. We monitor the model's latent trajectory in real-time — 14% better than classifiers.", comp: PillarReliable }, ]; return (

/ 03 — THE ML TRINITY THREE PILLARS · ONE STACK

The architecture of{' '} efficient {' '}intelligence.

India cannot win a brute-force hardware race. It must win the efficiency race. AttoFlow's stack rests on three pillars — each a deliberate refusal of waste.

{pillars.map((p, i) => (

{ e.currentTarget.style.transform = 'translateY(-4px)'; e.currentTarget.style.boxShadow = '0 24px 48px -24px rgba(10,10,10,0.18)'; }} onMouseLeave={e => { e.currentTarget.style.transform = 'translateY(0)'; e.currentTarget.style.boxShadow = 'none'; }}>

{p.num} {p.kicker.toUpperCase()}

{p.name}

{p.body}

))}

); }; window.Trinity = Trinity;