aankitdas's picture
feat: add RAGAS eval
af2f8e1
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>RAGAS Evaluation - Document Intelligence RAG</title>
<style>
/* Theme toggle */
.theme-toggle {
position: fixed;
top: 20px;
right: 20px;
padding: 8px 14px;
background: var(--surface);
border: 1px solid var(--border-soft);
border-radius: var(--radius-md);
cursor: pointer;
font-size: 0.85rem;
color: var(--text-main);
z-index: 100;
transition: background 0.15s ease;
}
.theme-toggle:hover {
background: var(--surface-subtle);
}
/* Button group */
.button-group {
display: flex;
gap: 10px;
margin-bottom: 16px;
}
.btn-small {
padding: 8px 14px;
background: var(--accent);
color: white;
border: none;
border-radius: var(--radius-sm);
font-size: 0.85rem;
cursor: pointer;
transition: background 0.15s ease;
}
.btn-small:hover {
background: #1d4ed8;
}
.btn-small.secondary {
background: var(--surface-subtle);
color: var(--text-main);
border: 1px solid var(--border-soft);
}
.btn-small.secondary:hover {
background: var(--border-soft);
}
/* Same design tokens as index.html */
:root {
--surface: #ffffff;
--surface-subtle: #fafafa;
--bg-main: #f5f7fb;
--card-bg: #ffffff;
--accent: #2563eb;
--accent-soft: #eff6ff;
--text-main: #111827;
--text-muted: #6b7280;
--border-soft: #e5e7eb;
--success: #16a34a;
--error: #dc2626;
--warning: #f59e0b;
--radius-sm: 6px;
--radius-md: 10px;
--radius-lg: 14px;
}
[data-theme="dark"] {
--surface: #1f2937;
--surface-subtle: #111827;
--bg-main: #0f172a;
--card-bg: #1e293b;
--accent: #60a5fa;
--accent-soft: #1e3a5f;
--text-main: #f1f5f9;
--text-muted: #94a3b8;
--border-soft: #334155;
}
* {
margin: 0;
padding: 0;
box-sizing: border-box;
}
body {
font-family: Inter, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, Helvetica, Arial, sans-serif;
background: var(--bg-main);
min-height: 100vh;
padding: 24px;
color: var(--text-main);
}
.container {
max-width: 900px;
margin: 0 auto;
}
header {
text-align: center;
margin-bottom: 36px;
}
header h1 {
font-size: 2rem;
font-weight: 600;
letter-spacing: -0.02em;
margin-bottom: 8px;
}
header p {
font-size: 1rem;
color: var(--text-muted);
margin-bottom: 16px;
}
.nav-links {
display: flex;
justify-content: center;
gap: 12px;
}
.nav-links a {
color: var(--accent);
text-decoration: none;
font-size: 0.9rem;
}
.nav-links a:hover {
text-decoration: underline;
}
.card {
background: var(--surface);
border-radius: var(--radius-lg);
padding: 28px;
border: 1px solid var(--border-soft);
box-shadow: 0 8px 20px rgba(0, 0, 0, 0.05);
margin-bottom: 24px;
}
.card h2 {
font-size: 1.2rem;
font-weight: 600;
margin-bottom: 18px;
}
.query-section {
display: flex;
gap: 12px;
margin-bottom: 16px;
}
.query-section input {
flex: 1;
padding: 12px 14px;
border: 1.5px solid var(--border-soft);
border-radius: var(--radius-md);
font-size: 0.95rem;
background: var(--surface);
color: var(--text-main);
}
.query-section input:focus {
outline: none;
border-color: var(--accent);
box-shadow: 0 0 0 3px var(--accent-soft);
}
.btn {
background: var(--accent);
color: white;
border: none;
padding: 12px 24px;
border-radius: var(--radius-md);
font-size: 0.95rem;
font-weight: 500;
cursor: pointer;
transition: background 0.15s ease, transform 0.15s ease;
}
.btn:hover {
background: #1d4ed8;
transform: translateY(-1px);
}
.btn:disabled {
background: var(--text-muted);
cursor: not-allowed;
transform: none;
}
.status {
padding: 12px;
border-radius: var(--radius-md);
font-size: 0.9rem;
margin-bottom: 16px;
}
.status.loading {
background: var(--accent-soft);
color: var(--accent);
}
.status.error {
background: #fef2f2;
color: var(--error);
}
.status.success {
background: #f0fdf4;
color: var(--success);
}
.hidden {
display: none;
}
/* Answer Section */
.answer-box {
background: var(--surface-subtle);
border-radius: var(--radius-md);
padding: 16px;
margin-bottom: 20px;
border-left: 4px solid var(--accent);
}
.answer-box p {
line-height: 1.6;
}
/* RAGAS Scores */
.scores-grid {
display: grid;
grid-template-columns: repeat(3, 1fr);
gap: 16px;
margin-bottom: 20px;
}
.score-card {
background: var(--surface-subtle);
border-radius: var(--radius-md);
padding: 20px;
text-align: center;
}
.score-label {
font-size: 0.85rem;
color: var(--text-muted);
margin-bottom: 8px;
text-transform: uppercase;
letter-spacing: 0.5px;
}
.score-value {
font-size: 2rem;
font-weight: 700;
}
.score-value.high {
color: var(--success);
}
.score-value.medium {
color: var(--warning);
}
.score-value.low {
color: var(--error);
}
/* Progress Bar */
.progress-bar {
height: 8px;
background: var(--border-soft);
border-radius: 4px;
overflow: hidden;
margin-top: 8px;
}
.progress-fill {
height: 100%;
border-radius: 4px;
transition: width 0.5s ease;
}
.progress-fill.high {
background: var(--success);
}
.progress-fill.medium {
background: var(--warning);
}
.progress-fill.low {
background: var(--error);
}
/* Aggregate Stats */
.aggregate-stats {
display: flex;
justify-content: space-around;
padding: 16px;
background: var(--surface-subtle);
border-radius: var(--radius-md);
}
.stat-item {
text-align: center;
}
.stat-value {
font-size: 1.5rem;
font-weight: 600;
color: var(--accent);
}
.stat-label {
font-size: 0.8rem;
color: var(--text-muted);
}
/* Sources */
.sources-section {
margin-top: 20px;
}
.sources-section h3 {
font-size: 1rem;
margin-bottom: 12px;
color: var(--text-muted);
}
.source-item {
background: var(--surface-subtle);
border-radius: var(--radius-sm);
padding: 12px;
margin-bottom: 8px;
font-size: 0.85rem;
}
.source-similarity {
color: var(--accent);
font-weight: 500;
margin-bottom: 4px;
}
/* Info Box */
.info-box {
background: var(--accent-soft);
border-left: 4px solid var(--accent);
padding: 14px;
border-radius: var(--radius-sm);
margin-bottom: 20px;
}
.info-box p {
font-size: 0.9rem;
color: var(--text-muted);
}
/* Time display */
.time-info {
font-size: 0.85rem;
color: var(--text-muted);
text-align: right;
margin-top: 12px;
}
@media (max-width: 600px) {
.scores-grid {
grid-template-columns: 1fr;
}
.query-section {
flex-direction: column;
}
}
</style>
</head>
<body>
<button class="theme-toggle" id="themeToggle">๐ŸŒ™ Dark mode</button>
<div class="container">
<header>
<h1>๐Ÿ”ฌ RAGAS Evaluation</h1>
<p>Test the RAG pipeline and see real-time quality metrics</p>
<div class="nav-links">
<a href="/">โ† Back to Main</a>
<a href="/evaluation">View Full Dashboard</a>
</div>
</header>
<!-- Query Section -->
<div class="card">
<h2>Ask a Question</h2>
<div class="info-box">
<p>Enter a question to query the RAG system. RAGAS will evaluate the response for
<strong>Faithfulness</strong> (is the answer grounded in context?) and <strong>Context
Precision</strong> (were the retrieved chunks useful?).
</p>
</div>
<div class="query-section">
<input type="text" id="queryInput" placeholder="e.g., What is machine learning?"
onkeypress="if(event.key === 'Enter') runEvaluation()">
<button class="btn" id="evalBtn" onclick="runEvaluation()">
๐Ÿ” Query & Evaluate
</button>
</div>
<div id="status" class="status hidden"></div>
</div>
<!-- Results Section -->
<div id="resultsSection" class="card hidden">
<h2>๐Ÿ“Š RAGAS Evaluation Results</h2>
<!-- Scores -->
<div class="scores-grid">
<div class="score-card">
<div class="score-label">Faithfulness</div>
<div class="score-value" id="faithScore">-</div>
<div class="progress-bar">
<div class="progress-fill" id="faithBar" style="width: 0%"></div>
</div>
</div>
<div class="score-card">
<div class="score-label">Context Precision</div>
<div class="score-value" id="precisionScore">-</div>
<div class="progress-bar">
<div class="progress-fill" id="precisionBar" style="width: 0%"></div>
</div>
</div>
<div class="score-card">
<div class="score-label">RAGAS Score</div>
<div class="score-value" id="ragasScore">-</div>
<div class="progress-bar">
<div class="progress-fill" id="ragasBar" style="width: 0%"></div>
</div>
</div>
</div>
<!-- Answer -->
<h3 style="margin-bottom: 12px;">๐Ÿ“ Generated Answer</h3>
<div class="answer-box">
<p id="answerText">-</p>
</div>
<!-- Sources -->
<div class="sources-section">
<h3>๐Ÿ“š Retrieved Sources</h3>
<div id="sourcesList"></div>
</div>
<div class="time-info">
<span id="responseTime"></span> | <span id="evalTime"></span>
</div>
</div>
<!-- Aggregate Stats -->
<div class="card">
<h2>๐Ÿ“ˆ Aggregate Statistics</h2>
<div class="button-group">
<button class="btn-small" onclick="loadAggregateStats()">๐Ÿ”„ Refresh</button>
<button class="btn-small secondary" onclick="clearRagasResults()">๐Ÿ—‘๏ธ Clear Results</button>
</div>
<div class="aggregate-stats">
<div class="stat-item">
<div class="stat-value" id="totalEvals">0</div>
<div class="stat-label">Total Evaluations</div>
</div>
<div class="stat-item">
<div class="stat-value" id="avgFaith">-</div>
<div class="stat-label">Avg Faithfulness</div>
</div>
<div class="stat-item">
<div class="stat-value" id="avgPrecision">-</div>
<div class="stat-label">Avg Precision</div>
</div>
<div class="stat-item">
<div class="stat-value" id="avgRagas">-</div>
<div class="stat-label">Avg RAGAS Score</div>
</div>
</div>
</div>
</div>
<script>
async function clearRagasResults() {
if (!confirm('Clear all RAGAS evaluation results? This cannot be undone.')) {
return;
}
try {
const response = await fetch(`${API_URL}/ragas/reset`, {
method: 'POST'
});
if (response.ok) {
// Reset displayed values
document.getElementById('totalEvals').textContent = '0';
document.getElementById('avgFaith').textContent = '-';
document.getElementById('avgPrecision').textContent = '-';
document.getElementById('avgRagas').textContent = '-';
alert('RAGAS results cleared!');
} else {
alert('Failed to clear results');
}
} catch (error) {
alert('Error: ' + error.message);
}
}
// Dark mode toggle
const themeToggle = document.getElementById("themeToggle");
const root = document.documentElement;
const savedTheme = localStorage.getItem("theme");
const prefersDark = window.matchMedia("(prefers-color-scheme: dark)").matches;
if (savedTheme) {
root.setAttribute("data-theme", savedTheme);
} else if (prefersDark) {
root.setAttribute("data-theme", "dark");
}
function updateToggleText() {
const isDark = root.getAttribute("data-theme") === "dark";
themeToggle.textContent = isDark ? "โ˜€๏ธ Light mode" : "๐ŸŒ™ Dark mode";
}
updateToggleText();
themeToggle.addEventListener("click", () => {
const isDark = root.getAttribute("data-theme") === "dark";
const newTheme = isDark ? "light" : "dark";
root.setAttribute("data-theme", newTheme);
localStorage.setItem("theme", newTheme);
updateToggleText();
});
const API_URL = window.location.origin;
async function runEvaluation() {
const query = document.getElementById('queryInput').value.trim();
if (!query) {
showStatus('Please enter a question', 'error');
return;
}
const btn = document.getElementById('evalBtn');
btn.disabled = true;
btn.textContent = 'โณ Evaluating...';
showStatus('Querying RAG and running RAGAS evaluation...', 'loading');
document.getElementById('resultsSection').classList.add('hidden');
try {
const response = await fetch(`${API_URL}/ragas/query-and-evaluate`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ query, top_k: 3 })
});
if (!response.ok) {
throw new Error('Evaluation failed');
}
const data = await response.json();
displayResults(data);
hideStatus();
loadAggregateStats();
} catch (error) {
showStatus(`Error: ${error.message}`, 'error');
} finally {
btn.disabled = false;
btn.textContent = '๐Ÿ” Query & Evaluate';
}
}
function displayResults(data) {
document.getElementById('resultsSection').classList.remove('hidden');
// Answer
document.getElementById('answerText').textContent = data.answer;
// RAGAS Scores
if (data.ragas) {
setScore('faithScore', 'faithBar', data.ragas.faithfulness);
setScore('precisionScore', 'precisionBar', data.ragas.context_precision);
setScore('ragasScore', 'ragasBar', data.ragas.ragas_score);
document.getElementById('evalTime').textContent =
`Eval: ${data.ragas.eval_time_ms.toFixed(0)}ms`;
}
document.getElementById('responseTime').textContent =
`Response: ${data.response_time_ms.toFixed(0)}ms`;
// Sources
const sourcesList = document.getElementById('sourcesList');
sourcesList.innerHTML = data.sources.map(source => `
<div class="source-item">
<div class="source-similarity">๐Ÿ“Œ Similarity: ${(source.similarity * 100).toFixed(0)}%</div>
<div>${source.preview}</div>
</div>
`).join('');
}
function setScore(valueId, barId, score) {
const valueEl = document.getElementById(valueId);
const barEl = document.getElementById(barId);
const percent = (score * 100).toFixed(0);
valueEl.textContent = percent + '%';
barEl.style.width = percent + '%';
// Color coding
let colorClass = 'high';
if (score < 0.7) colorClass = 'medium';
if (score < 0.5) colorClass = 'low';
valueEl.className = 'score-value ' + colorClass;
barEl.className = 'progress-fill ' + colorClass;
}
async function loadAggregateStats() {
try {
const response = await fetch(`${API_URL}/ragas/metrics`);
const data = await response.json();
document.getElementById('totalEvals').textContent = data.total_evaluations || 0;
document.getElementById('avgFaith').textContent =
data.avg_faithfulness ? (data.avg_faithfulness * 100).toFixed(0) + '%' : '-';
document.getElementById('avgPrecision').textContent =
data.avg_context_precision ? (data.avg_context_precision * 100).toFixed(0) + '%' : '-';
document.getElementById('avgRagas').textContent =
data.avg_ragas_score ? (data.avg_ragas_score * 100).toFixed(0) + '%' : '-';
} catch (e) {
console.error('Failed to load aggregate stats:', e);
}
}
function showStatus(message, type) {
const status = document.getElementById('status');
status.textContent = message;
status.className = `status ${type}`;
status.classList.remove('hidden');
}
function hideStatus() {
document.getElementById('status').classList.add('hidden');
}
// Load stats on page load
window.addEventListener('load', loadAggregateStats);
</script>
</body>
</html>