index.html

<!doctype html>
<html lang="en">
<head>
    <meta charset="UTF-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
    <title>KR Labs: grounded AI</title>
    <meta name="description" content="KR Labs builds open-source, transparent NLP technology: hallucination detection, verbatim RAG, and rule-based models. A spin-off of TU Wien." />
    <link rel="preconnect" href="https://fonts.googleapis.com" />
    <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
    <link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600&family=Space+Grotesk:wght@600;700&family=Space+Mono:wght@700&display=swap" rel="stylesheet" />
    <style>
        :root {
            --ink: #0F0F0F;
            --paper: #FFFFFF;
            --graphite: #2A2D33;
            --slate: #4A5060;
            --mist: #9CA3AF;
            --signal: #FF3D5A;
            --signal-tint: rgb(255 61 90 / 0.12);
            --rule: #E5E7EB;
        }
        @media (prefers-color-scheme: dark) {
            :root {
                --ink: #F5F6F8;
                --paper: #0F1115;
                --graphite: #D1D5DB;
                --slate: #8B92A0;
                --mist: #4A5060;
                --signal: #FF5A72;
                --signal-tint: rgb(255 90 114 / 0.16);
                --rule: #23262D;
            }
        }
        * { box-sizing: border-box; margin: 0; padding: 0; }
        body {
            font-family: "Inter", -apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica, Arial, sans-serif;
            font-size: 17px;
            line-height: 1.6;
            color: var(--ink);
            background-color: var(--paper);
        }
        a { color: inherit; }
        .wrap { max-width: 880px; margin: 0 auto; padding: 0 24px; }

        .mono { font-family: "Space Mono", Menlo, monospace; font-weight: 700; }
        .bracket { color: var(--signal); }

        header {
            display: flex;
            align-items: baseline;
            justify-content: space-between;
            padding: 28px 0;
            flex-wrap: wrap;
            gap: 12px;
        }
        .mark { font-size: 20px; text-decoration: none; }
        .mark .labs { font-family: "Space Grotesk", sans-serif; font-weight: 700; }
        nav a {
            font-size: 15px;
            font-weight: 500;
            color: var(--slate);
            text-decoration: none;
            margin-left: 20px;
        }
        nav a:hover { color: var(--ink); box-shadow: 0 1px 0 var(--signal); }

        .hero { padding: 72px 0 56px; }
        .hero h1 {
            font-family: "Space Grotesk", sans-serif;
            font-weight: 700;
            font-size: clamp(32px, 5.5vw, 48px);
            line-height: 1.1;
            letter-spacing: -0.015em;
            max-width: 18ch;
        }
        .hero p {
            margin-top: 20px;
            font-size: 19px;
            color: var(--slate);
            max-width: 60ch;
        }
        .hero p a { color: var(--ink); text-decoration: none; box-shadow: 0 1px 0 var(--signal); }

        h2 {
            font-family: "Space Grotesk", sans-serif;
            font-weight: 600;
            font-size: 22px;
            letter-spacing: -0.015em;
            margin-bottom: 8px;
        }
        .section { padding: 32px 0; }
        .section > .lede { color: var(--slate); margin-bottom: 28px; max-width: 65ch; }
        .section > .lede a { color: var(--ink); text-decoration: none; box-shadow: 0 1px 0 var(--signal); }

        .project {
            border-top: 1px solid var(--rule);
            padding: 32px 0;
        }
        .project h3 { font-size: 20px; margin-bottom: 2px; }
        .project .name a { text-decoration: none; }
        .project .name a:hover .word { box-shadow: 0 1px 0 var(--signal); }
        .project .tag {
            font-family: "Space Grotesk", sans-serif;
            font-weight: 600;
            font-size: 17px;
            color: var(--graphite);
            margin: 6px 0 10px;
        }
        .project p.desc { color: var(--slate); max-width: 68ch; }
        .links { margin-top: 14px; font-size: 15px; font-weight: 500; }
        .links a { color: var(--ink); text-decoration: none; box-shadow: 0 1px 0 var(--signal); }
        .links span.sep { color: var(--mist); margin: 0 10px; }

        .more { list-style: none; }
        .more li { margin: 10px 0; color: var(--slate); }
        .more a { color: var(--ink); text-decoration: none; box-shadow: 0 1px 0 var(--signal); }

        footer {
            border-top: 1px solid var(--rule);
            margin-top: 48px;
            padding: 32px 0 48px;
            font-size: 15px;
            color: var(--slate);
            display: flex;
            justify-content: space-between;
            flex-wrap: wrap;
            gap: 12px;
        }
        footer a { color: var(--ink); text-decoration: none; box-shadow: 0 1px 0 var(--signal); }
    </style>
</head>
<body>
    <div class="wrap">
        <header>
            <a class="mark mono" href="/"><span class="bracket">[</span>KR<span class="bracket">]</span> <span class="labs">Labs</span></a>
            <nav>
                <a href="https://github.com/KRLabsOrg">GitHub</a>
                <a href="https://huggingface.co/KRLabsOrg">Hugging Face</a>
                <a href="https://krlabs.eu">krlabs.eu</a>
                <a href="mailto:contact@krlabs.eu">Contact</a>
            </nav>
        </header>

        <section class="hero">
            <h1>Grounded AI for teams that cannot afford made-up answers.</h1>
            <p>KR Labs is a spin-off of <a href="https://informatics.tuwien.ac.at/">TU Wien</a> building open-source, transparent NLP technology. We publish what we ship, and ship what we publish.</p>
        </section>

        <section class="section" id="projects">
            <h2>Open-source projects</h2>
            <p class="lede">Everything below is on <a href="https://github.com/KRLabsOrg">GitHub</a>, with models and datasets on Hugging Face.</p>

            <div class="project">
                <h3 class="name mono"><a href="https://github.com/KRLabsOrg/LettuceDetect"><span class="bracket">[</span><span class="word">lettucedetect</span><span class="bracket">]</span></a></h3>
                <p class="tag">Hallucination detection for AI outputs. Flags ungrounded claims before they ship.</p>
                <p class="desc">Span-level detection of unsupported content in RAG answers, generated code, and agentic tool output. Compact encoder models that outperform prompt-based judges at a fraction of the cost, in seven languages.</p>
                <p class="links">
                    <a href="https://github.com/KRLabsOrg/LettuceDetect">GitHub</a><span class="sep">·</span>
                    <a href="https://krlabsorg.github.io/LettuceDetect/">Docs</a><span class="sep">·</span>
                    <a href="https://pypi.org/project/lettucedetect/">PyPI</a><span class="sep">·</span>
                    <a href="https://huggingface.co/KRLabsOrg">Models</a><span class="sep">·</span>
                    <a href="https://arxiv.org/abs/2502.17125">Paper</a>
                </p>
            </div>

            <div class="project">
                <h3 class="name mono"><a href="https://github.com/KRLabsOrg/verbatim-rag"><span class="bracket">[</span><span class="word">verbatim</span><span class="bracket">]</span></a></h3>
                <p class="tag">Retrieval-augmented generation with span-level attribution.</p>
                <p class="desc">VerbatimRAG prevents hallucination by construction: extractive models select the relevant source spans, and the LLM only fills templates around them, so every part of an answer traces back to a verbatim source span.</p>
                <p class="links">
                    <a href="https://github.com/KRLabsOrg/verbatim-rag">GitHub</a><span class="sep">·</span>
                    <a href="https://github.com/KRLabsOrg/acl-verbatim">ACL Anthology demo</a>
                </p>
            </div>

            <div class="project">
                <h3 class="name mono"><a href="https://github.com/KRLabsOrg/rulechef"><span class="bracket">[</span><span class="word">rulechef</span><span class="bracket">]</span></a></h3>
                <p class="tag">Learn rule-based models from examples using LLM-powered synthesis.</p>
                <p class="desc">You provide labeled examples, RuleChef generates regex, code, or spaCy rules, and those rules run locally with no LLM at inference time: fast, deterministic, and inspectable.</p>
                <p class="links">
                    <a href="https://github.com/KRLabsOrg/rulechef">GitHub</a><span class="sep">·</span>
                    <a href="https://krlabsorg.github.io/rulechef/">Docs</a><span class="sep">·</span>
                    <a href="https://pypi.org/project/rulechef/">PyPI</a>
                </p>
            </div>
        </section>

        <section class="section" id="more">
            <h2>Also on our GitHub</h2>
            <ul class="more">
                <li><a href="https://github.com/KRLabsOrg/squeez">squeez</a>: squeeze verbose LLM agent tool output down to only the relevant lines</li>
                <li><a href="https://github.com/KRLabsOrg/RAGFactChecker">RAGFactChecker</a>: triplet-based fact checking for RAG answers</li>
                <li><a href="https://github.com/KRLabsOrg/verbatim-mcp">verbatim-mcp</a>: MCP server for querying academic papers via VerbatimRAG</li>
            </ul>
        </section>

        <footer>
            <span>KR Labs GmbH, Vienna · <a href="mailto:contact@krlabs.eu">contact@krlabs.eu</a></span>
            <span><a href="imprint.html">Imprint</a></span>
        </footer>
    </div>
</body>
</html>