Blame for docs/cortex-onepager.html

<!doctype html>
<html><head><meta charset="utf-8">
<style>
:root{--ink:#0f172a;--sub:#475569;--bg:#f8fafc;--line:#e2e8f0;--or:#F97316;--or2:#fb923c;--navy:#0B1B3A;--blue:#1E3A8A;--teal:#0EA5A4;--grey:#94a3b8}
*{box-sizing:border-box;margin:0;padding:0}
body{width:1200px;background:var(--bg);font-family:Inter,"Liberation Sans","DejaVu Sans",sans-serif;color:var(--ink)}
.hero{background:linear-gradient(135deg,#0b1b3a 0%,#132c5c 65%,#1e3a8a 100%);color:#fff;padding:48px 64px 44px;display:flex;gap:36px;align-items:center}
.hero img{width:150px;height:150px;background:#fff;border-radius:28px;padding:10px;flex:none}
.tag{display:inline-block;background:var(--or);color:#fff;font-weight:800;font-size:14px;letter-spacing:.14em;padding:6px 14px;border-radius:999px;text-transform:uppercase}
h1{font-size:54px;font-weight:900;line-height:1.05;margin:14px 0 12px;letter-spacing:-.02em}h1 span{color:var(--or2)}
.lead{font-size:21px;line-height:1.45;color:#cbd5e1}.lead b{color:#fff}
.wrap{padding:36px 64px 40px}
h2{font-size:15px;letter-spacing:.16em;text-transform:uppercase;color:var(--or);font-weight:800;margin-bottom:14px}
.sec{margin-top:32px}.sec:first-child{margin-top:0}
.row{display:grid;gap:18px}.c2{grid-template-columns:1fr 1fr}.c3{grid-template-columns:repeat(3,1fr)}.c4{grid-template-columns:repeat(4,1fr)}
.card{background:#fff;border:1px solid var(--line);border-radius:18px;padding:22px}
.vs .card h3{font-size:21px;font-weight:800;margin:6px 0 8px}.vs .card p{font-size:16px;color:var(--sub);line-height:1.45}
.pill{font-size:13px;font-weight:800;letter-spacing:.1em;text-transform:uppercase;color:var(--grey)}.good{border:2px solid var(--or)}.good .pill{color:var(--or)}
.res{display:grid;grid-template-columns:1.25fr 1fr;gap:18px}
.bars .lbl{display:flex;justify-content:space-between;font-size:15px;font-weight:700;margin:14px 0 6px}.bars .lbl:first-of-type{margin-top:4px}
.bar{height:22px;border-radius:7px;background:#e2e8f0;overflow:hidden}.bar i{display:block;height:100%}
.v{background:#94a3b8}.g{background:linear-gradient(90deg,var(--or),var(--or2))}
.metric{font-size:13px;color:var(--sub);margin-top:16px;font-weight:700;letter-spacing:.06em;text-transform:uppercase}
.big{display:flex;flex-direction:column;justify-content:center;gap:14px}
.kpi{background:#fff7ed;border-radius:14px;padding:16px 18px}.kpi .n{font-size:40px;font-weight:900;color:var(--navy);letter-spacing:-.02em}.kpi .n small{font-size:20px;color:var(--grey);font-weight:800}
.kpi .l{font-size:14px;color:var(--sub);margin-top:2px;line-height:1.35}
.more .card{text-align:left}.more .n{font-size:30px;font-weight:900;color:var(--navy)}.more .t{font-size:16px;font-weight:800;margin:6px 0 4px}.more .d{font-size:13.5px;color:var(--sub);line-height:1.4}
.checks li{list-style:none;font-size:15.5px;line-height:1.4;padding:7px 0 7px 32px;position:relative}
.checks li:before{content:"";position:absolute;left:0;top:9px;width:20px;height:20px;border-radius:50%;background:var(--teal)}
.checks li:after{content:"";position:absolute;left:7px;top:13px;width:5px;height:10px;border:solid #fff;border-width:0 3px 3px 0;transform:rotate(45deg)}
.note{font-size:12.5px;color:var(--grey);margin-top:10px;line-height:1.4}
.foot{background:var(--navy);color:#fff;padding:26px 64px;display:grid;grid-template-columns:1.6fr 1fr 1fr 1fr;gap:20px;align-items:center}
.foot .k{font-size:12px;letter-spacing:.14em;text-transform:uppercase;color:var(--or2);font-weight:800}.foot .fv{font-size:19px;font-weight:800;margin-top:5px}.foot .s{font-size:13px;color:#94a3b8;margin-top:2px}
</style></head><body>
<div class="hero">
  <img src="brand/cortex-logo.svg" alt="">
  <div>
    <span class="tag">Cortex · Live demo</span>
    <h1>Why <span>FalkorDB</span> for AI<br>engagements, and more</h1>
    <p class="lead">RAG finds <b>similar text</b>. Real questions need <b>connected facts</b>. Same 73 bank PDFs, same LLM, same questions: the only difference is the <b>graph</b>.</p>
  </div>
</div>
<div class="wrap">
  <div class="sec">
    <h2>Measured: vector RAG vs GraphRAG on FalkorDB (34 questions, known answers)</h2>
    <div class="res">
      <div class="card bars">
        <div class="metric">Correct IDs found (average recall)</div>
        <div class="lbl"><span>Vector RAG</span><span>70%</span></div><div class="bar"><i class="v" style="width:70%"></i></div>
        <div class="lbl"><span>GraphRAG on FalkorDB</span><span style="color:var(--or)">97%</span></div><div class="bar"><i class="g" style="width:97%"></i></div>
        <div class="metric">Multi-hop questions (policy → control → procedure → finding)</div>
        <div class="lbl"><span>Vector RAG</span><span>47%</span></div><div class="bar"><i class="v" style="width:47%"></i></div>
        <div class="lbl"><span>GraphRAG on FalkorDB</span><span style="color:var(--or)">97%</span></div><div class="bar"><i class="g" style="width:97%"></i></div>
        <p class="note">Simple single-fact and list questions: both 100%. The graph wins where facts must be connected.</p>
      </div>
      <div class="big">
        <div class="kpi"><div class="n">31<small>/34</small> vs 19<small>/34</small></div><div class="l">fully correct answers: GraphRAG vs vector RAG</div></div>
        <div class="kpi"><div class="n">≈ same</div><div class="l">tokens per correct answer (2,535 vs 2,540): better answers without higher cost</div></div>
        <div class="kpi"><div class="n">$0</div><div class="l">run cost: free Gemini tier, local embeddings, free hosting</div></div>
      </div>
    </div>
  </div>

  <div class="sec">
    <h2>…and more: what the same graph does beyond RAG</h2>
    <div class="row c4 more">
      <div class="card"><div class="n">60/60</div><div class="t">Fraud rings caught</div><div class="d">Across 1,005,040 nodes in under 2 s. Zero false alarms.</div></div>
      <div class="card"><div class="n">0.4 ms</div><div class="t">Regulatory impact</div><div class="d">Regulation → policies → controls → procedures → teams → KRIs → findings.</div></div>
      <div class="card"><div class="n">7 roles</div><div class="t">Access-aware AI</div><div class="d">A teller and a CRO get different answers. Restricted sources never reach the LLM.</div></div>
      <div class="card"><div class="n">1 per user</div><div class="t">Private sandboxes</div><div class="d">Upload a PDF, get your own isolated graph, wiped after 24 hours.</div></div>
    </div>
  </div>

  <div class="sec row c2">
    <div class="card"><h2>Why you can trust it</h2><ul class="checks">
      <li><b>Fair test:</b> same PDFs, chunks, embeddings, LLM and prompt</li>
      <li><b>Known answers:</b> scored against the bank's source data</li>
      <li><b>Honest:</b> both approaches tie on simple lookups; we show it</li>
      <li><b>Open:</b> Cypher, pipeline and code visible to engineers</li>
      <li><b>Zero risk:</b> 100% synthetic "Cortex Bank" (est. 1926)</li></ul></div>
    <div class="card"><h2>Try it yourself</h2><ul class="checks">
      <li><b>Ask Cortex:</b> both answers side by side, with sources and tokenomics</li>
      <li><b>Switch roles:</b> compare what a teller and a CRO can see</li>
      <li><b>One-click scenarios:</b> fraud, impact, risk roll-up, lineage</li>
      <li><b>Upload your PDF</b> into a private sandbox graph</li>
      <li><b>Engineer view:</b> run the evaluation, see every query</li></ul>
      <p class="note">Results: 34-question run on 30 Sep 2026, Gemini Flash-Lite (free tier), role CRO. Synthetic data; your results may differ.</p></div>
  </div>
</div>
<div class="foot">
  <div><div class="k">Live demo</div><div class="fv">sk-aiu-cortex.hf.space</div><div class="s">Access PIN from the demo owner</div></div>
  <div><div class="k">Stack</div><div class="fv">FalkorDB + LangChain</div><div class="s">Embedded, one container</div></div>
  <div><div class="k">LLM</div><div class="fv">Gemini (free)</div><div class="s">or your own Gemini / Claude key</div></div>
  <div><div class="k">Cost</div><div class="fv">$0</div><div class="s">Free tiers end to end</div></div>
</div>
</body></html>