RAG v2: HNSW ANN, cross-encoder reranking, evaluation logging, analytics dashboard

analytics.py

@@ -0,0 +1,81 @@
+import json
+from collections import defaultdict
+from datetime import datetime
+
+LOG_FILE = "rag_eval_logs.jsonl"
+
+def get_analytics():
+    """Parse logs and return analytics data."""
+    total = 0
+    known_count = 0
+    unknown_count = 0
+    conf_sum = 0.0
+    queries = []
+    unknown_queries = []
+    
+    try:
+        with open(LOG_FILE, "r", encoding="utf-8") as f:
+            for line in f:
+                line = line.strip()
+                if not line:
+                    continue
+                
+                total += 1
+                data = json.loads(line)
+                
+                if data.get("answer_known"):
+                    known_count += 1
+                else:
+                    unknown_count += 1
+                    unknown_queries.append({
+                        "query": data.get("query"),
+                        "timestamp": datetime.fromtimestamp(data.get("timestamp", 0)).strftime("%Y-%m-%d %H:%M")
+                    })
+                    
+                conf_sum += data.get("confidence", 0.0)
+                queries.append({
+                    "query": data.get("query"),
+                    "confidence": data.get("confidence", 0.0),
+                    "answer_known": data.get("answer_known", False)
+                })
+        
+        if total == 0:
+            return {
+                "total_queries": 0,
+                "knowledge_rate": 0,
+                "avg_confidence": 0,
+                "known_count": 0,
+                "unknown_count": 0,
+                "recent_unknown": [],
+                "top_queries": []
+            }
+        
+        knowledge_rate = (known_count / total) * 100
+        avg_confidence = conf_sum / total
+        
+        # Get top 10 most recent queries
+        top_queries = queries[-10:][::-1]  # Last 10, reversed
+        
+        # Get recent unknown queries (last 5)
+        recent_unknown = unknown_queries[-5:][::-1]
+        
+        return {
+            "total_queries": total,
+            "knowledge_rate": round(knowledge_rate, 1),
+            "avg_confidence": round(avg_confidence, 2),
+            "known_count": known_count,
+            "unknown_count": unknown_count,
+            "recent_unknown": recent_unknown,
+            "top_queries": top_queries
+        }
+    
+    except FileNotFoundError:
+        return {
+            "total_queries": 0,
+            "knowledge_rate": 0,
+            "avg_confidence": 0,
+            "known_count": 0,
+            "unknown_count": 0,
+            "recent_unknown": [],
+            "top_queries": []
+        }

analyze_logs.py

@@ -0,0 +1,47 @@
+import json
+from collections import Counter
+
+LOG_FILE = "rag_eval_logs.jsonl"
+
+def analyze():
+    print(f"--- Analyzing {LOG_FILE} ---\n")
+    
+    total = 0
+    known_count = 0
+    unknown_count = 0
+    conf_sum = 0.0
+
+    try:
+        with open(LOG_FILE, "r", encoding="utf-8") as f:
+            for line in f:
+                line = line.strip()
+                if not line: continue
+                
+                total += 1
+                data = json.loads(line)
+                
+                if data.get("answer_known"):
+                    known_count += 1
+                else:
+                    unknown_count += 1
+                    
+                conf_sum += data.get("confidence", 0.0)
+
+        if total == 0:
+            print("No logs found.")
+            return
+
+        print(f"Total Queries:      {total}")
+        print(f"Answered (Known):   {known_count}")
+        print(f"Unanswered (False): {unknown_count}")
+        print(f"Average Confidence: {conf_sum / total:.2f}")
+        print("-" * 30)
+        
+        accuracy = (known_count / total) * 100
+        print(f"System 'Knowledge Rate': {accuracy:.1f}%")
+
+    except FileNotFoundError:
+        print(f"Log file {LOG_FILE} not found.")
+
+if __name__ == "__main__":
+    analyze()

eval_logger.py

@@ -0,0 +1,21 @@
+import json
+from time import time
+
+LOG_FILE = "rag_eval_logs.jsonl"
+
+def log_eval(
+    query: str,
+    retrieved_count: int,
+    confidence: float,
+    answer_known: bool
+):
+    record = {
+        "timestamp": time(),
+        "query": query,
+        "retrieved_count": retrieved_count,
+        "confidence": confidence,
+        "answer_known": answer_known
+    }
+
+    with open(LOG_FILE, "a", encoding="utf-8") as f:
+        f.write(json.dumps(record) + "\n")

frontend/analytics.html

@@ -0,0 +1,334 @@
+<!DOCTYPE html>
+<html lang="en">
+
+<head>
+    <meta charset="UTF-8" />
+    <title>Analytics - Gemini RAG</title>
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap" rel="stylesheet">
+
+    <style>
+        :root {
+            --bg: radial-gradient(1200px 600px at top, #e0e7ff 0%, #f8fafc 60%);
+            --card: rgba(255, 255, 255, 0.9);
+            --border: rgba(15, 23, 42, 0.08);
+            --primary: #4f46e5;
+            --secondary: #0ea5e9;
+            --text: #0f172a;
+            --muted: #64748b;
+            --success: #16a34a;
+            --error: #dc2626;
+        }
+
+        [data-theme="dark"] {
+            --bg: radial-gradient(1200px 600px at top, #1e1b4b 0%, #0f172a 60%);
+            --card: rgba(30, 41, 59, 0.9);
+            --border: rgba(148, 163, 184, 0.1);
+            --primary: #818cf8;
+            --secondary: #38bdf8;
+            --text: #f1f5f9;
+            --muted: #94a3b8;
+            --success: #4ade80;
+            --error: #f87171;
+        }
+
+        * {
+            box-sizing: border-box;
+            font-family: Inter, sans-serif;
+        }
+
+        body {
+            margin: 0;
+            min-height: 100vh;
+            background: var(--bg);
+            padding: 40px 16px;
+            color: var(--text);
+            transition: background 0.3s ease, color 0.3s ease;
+        }
+
+        .container {
+            max-width: 1200px;
+            margin: 0 auto;
+        }
+
+        .header {
+            display: flex;
+            justify-content: space-between;
+            align-items: center;
+            margin-bottom: 32px;
+        }
+
+        h1 {
+            font-size: 2.2rem;
+            margin: 0;
+            font-weight: 700;
+            background: linear-gradient(135deg, #4f46e5, #06b6d4);
+            background-clip: text;
+            -webkit-background-clip: text;
+            -webkit-text-fill-color: transparent;
+        }
+
+        .back-btn {
+            padding: 10px 20px;
+            background: var(--primary);
+            color: white;
+            text-decoration: none;
+            border-radius: 12px;
+            font-weight: 600;
+            transition: transform 0.2s ease;
+        }
+
+        .back-btn:hover {
+            transform: translateY(-2px);
+        }
+
+        .stats-grid {
+            display: grid;
+            grid-template-columns: repeat(auto-fit, minmax(240px, 1fr));
+            gap: 20px;
+            margin-bottom: 32px;
+        }
+
+        .stat-card {
+            background: var(--card);
+            backdrop-filter: blur(16px);
+            border-radius: 18px;
+            padding: 24px;
+            border: 1px solid var(--border);
+            box-shadow: 0 4px 12px rgba(0, 0, 0, 0.05);
+        }
+
+        .stat-label {
+            font-size: 0.85rem;
+            color: var(--muted);
+            margin-bottom: 8px;
+            text-transform: uppercase;
+            letter-spacing: 0.5px;
+        }
+
+        .stat-value {
+            font-size: 2.5rem;
+            font-weight: 700;
+            color: var(--primary);
+        }
+
+        .card {
+            background: var(--card);
+            backdrop-filter: blur(16px);
+            border-radius: 18px;
+            padding: 28px;
+            border: 1px solid var(--border);
+            box-shadow: 0 4px 12px rgba(0, 0, 0, 0.05);
+            margin-bottom: 24px;
+        }
+
+        .card h2 {
+            margin-top: 0;
+            margin-bottom: 20px;
+            font-size: 1.3rem;
+        }
+
+        table {
+            width: 100%;
+            border-collapse: collapse;
+        }
+
+        th,
+        td {
+            text-align: left;
+            padding: 12px;
+            border-bottom: 1px solid var(--border);
+        }
+
+        th {
+            font-weight: 600;
+            color: var(--muted);
+            font-size: 0.85rem;
+            text-transform: uppercase;
+            letter-spacing: 0.5px;
+        }
+
+        .badge {
+            display: inline-block;
+            padding: 4px 10px;
+            border-radius: 12px;
+            font-size: 0.75rem;
+            font-weight: 600;
+        }
+
+        .badge-success {
+            background: #dcfce7;
+            color: #166534;
+        }
+
+        .badge-error {
+            background: #fee2e2;
+            color: #991b1b;
+        }
+
+        .theme-toggle {
+            position: fixed;
+            top: 20px;
+            right: 20px;
+            background: var(--card);
+            border: 1px solid var(--border);
+            border-radius: 12px;
+            padding: 10px;
+            cursor: pointer;
+            font-size: 1.4rem;
+            transition: transform 0.2s ease;
+            box-shadow: 0 4px 12px rgba(0, 0, 0, 0.1);
+        }
+
+        .theme-toggle:hover {
+            transform: scale(1.1);
+        }
+
+        .empty-state {
+            text-align: center;
+            padding: 60px 20px;
+            color: var(--muted);
+        }
+    </style>
+</head>
+
+<body>
+    <button class="theme-toggle" onclick="toggleTheme()" title="Toggle dark mode">🌙</button>
+
+    <div class="container">
+        <div class="header">
+            <h1>📊 Analytics Dashboard</h1>
+            <a href="/" class="back-btn">← Back to RAG</a>
+        </div>
+
+        <div id="stats-container">
+            <div class="empty-state">
+                <h2>Loading analytics...</h2>
+            </div>
+        </div>
+    </div>
+
+    <script>
+        // ===== THEME TOGGLE =====
+        function toggleTheme() {
+            const html = document.documentElement;
+            const currentTheme = html.getAttribute('data-theme');
+            const newTheme = currentTheme === 'dark' ? 'light' : 'dark';
+
+            html.setAttribute('data-theme', newTheme);
+            localStorage.setItem('theme', newTheme);
+
+            const btn = document.querySelector('.theme-toggle');
+            btn.textContent = newTheme === 'dark' ? '☀️' : '🌙';
+        }
+
+        // Load saved theme
+        (function () {
+            const savedTheme = localStorage.getItem('theme') || 'light';
+            document.documentElement.setAttribute('data-theme', savedTheme);
+            const btn = document.querySelector('.theme-toggle');
+            if (btn) btn.textContent = savedTheme === 'dark' ? '☀️' : '🌙';
+        })();
+
+        // ===== LOAD ANALYTICS =====
+        async function loadAnalytics() {
+            try {
+                const res = await fetch('/analytics');
+                const data = await res.json();
+
+                if (data.total_queries === 0) {
+                    document.getElementById('stats-container').innerHTML = `
+            <div class="empty-state">
+              <h2>No data yet</h2>
+              <p>Start asking questions to see analytics!</p>
+            </div>
+          `;
+                    return;
+                }
+
+                const html = `
+          <div class="stats-grid">
+            <div class="stat-card">
+              <div class="stat-label">Total Queries</div>
+              <div class="stat-value">${data.total_queries}</div>
+            </div>
+            <div class="stat-card">
+              <div class="stat-label">Knowledge Rate</div>
+              <div class="stat-value">${data.knowledge_rate}%</div>
+            </div>
+            <div class="stat-card">
+              <div class="stat-label">Avg Confidence</div>
+              <div class="stat-value">${(data.avg_confidence * 100).toFixed(0)}%</div>
+            </div>
+            <div class="stat-card">
+              <div class="stat-label">Unknown Queries</div>
+              <div class="stat-value" style="color: var(--error)">${data.unknown_count}</div>
+            </div>
+          </div>
+
+          <div class="card">
+            <h2>Recent Queries</h2>
+            <table>
+              <thead>
+                <tr>
+                  <th>Query</th>
+                  <th>Confidence</th>
+                  <th>Status</th>
+                </tr>
+              </thead>
+              <tbody>
+                ${data.top_queries.map(q => `
+                  <tr>
+                    <td>${q.query}</td>
+                    <td>${(q.confidence * 100).toFixed(0)}%</td>
+                    <td>
+                      <span class="badge ${q.answer_known ? 'badge-success' : 'badge-error'}">
+                        ${q.answer_known ? 'Known' : 'Unknown'}
+                      </span>
+                    </td>
+                  </tr>
+                `).join('')}
+              </tbody>
+            </table>
+          </div>
+
+          ${data.recent_unknown.length > 0 ? `
+            <div class="card">
+              <h2>Recent "I Don't Know" Queries</h2>
+              <table>
+                <thead>
+                  <tr>
+                    <th>Query</th>
+                    <th>Time</th>
+                  </tr>
+                </thead>
+                <tbody>
+                  ${data.recent_unknown.map(q => `
+                    <tr>
+                      <td>${q.query}</td>
+                      <td>${q.timestamp}</td>
+                    </tr>
+                  `).join('')}
+                </tbody>
+              </table>
+            </div>
+          ` : ''}
+        `;
+
+                document.getElementById('stats-container').innerHTML = html;
+            } catch (e) {
+                document.getElementById('stats-container').innerHTML = `
+          <div class="empty-state">
+            <h2>Error loading analytics</h2>
+            <p>${e.message}</p>
+          </div>
+        `;
+            }
+        }
+
+        // Load on page load
+        loadAnalytics();
+    </script>
+</body>
+
+</html>

frontend/index.html

@@ -1,5 +1,6 @@
 <!DOCTYPE html>
 <html lang="en">
+
 <head>
   <meta charset="UTF-8" />
   <title>Gemini RAG Assistant</title>
@@ -10,8 +11,8 @@
   <style>
     :root {
       --bg: radial-gradient(1200px 600px at top, #e0e7ff 0%, #f8fafc 60%);
-      --card: rgba(255,255,255,0.9);
-      --border: rgba(15,23,42,0.08);
+      --card: rgba(255, 255, 255, 0.9);
+      --border: rgba(15, 23, 42, 0.08);
       --primary: #4f46e5;
       --secondary: #0ea5e9;
       --text: #0f172a;
@@ -20,7 +21,22 @@
       --success: #16a34a;
     }
 
-    * { box-sizing: border-box; font-family: Inter, sans-serif; }
+    [data-theme="dark"] {
+      --bg: radial-gradient(1200px 600px at top, #1e1b4b 0%, #0f172a 60%);
+      --card: rgba(30, 41, 59, 0.9);
+      --border: rgba(148, 163, 184, 0.1);
+      --primary: #818cf8;
+      --secondary: #38bdf8;
+      --text: #f1f5f9;
+      --muted: #94a3b8;
+      --error: #f87171;
+      --success: #4ade80;
+    }
+
+    * {
+      box-sizing: border-box;
+      font-family: Inter, sans-serif;
+    }
 
     body {
       margin: 0;
@@ -30,6 +46,7 @@
       justify-content: center;
       padding: 40px 16px;
       color: var(--text);
+      transition: background 0.3s ease, color 0.3s ease;
     }
 
     .container {
@@ -40,7 +57,7 @@
       border-radius: 24px;
       padding: 36px;
       border: 1px solid var(--border);
-      box-shadow: 0 40px 120px rgba(15,23,42,.15);
+      box-shadow: 0 40px 120px rgba(15, 23, 42, .15);
     }
 
     h1 {
@@ -48,6 +65,7 @@
       margin: 0;
       font-weight: 700;
       background: linear-gradient(135deg, #4f46e5, #06b6d4);
+      background-clip: text;
       -webkit-background-clip: text;
       -webkit-text-fill-color: transparent;
     }
@@ -60,7 +78,7 @@
 
     .card {
       margin-top: 28px;
-      background: white;
+      background: var(--card);
       border-radius: 18px;
       padding: 24px;
       border: 1px solid var(--border);
@@ -72,13 +90,15 @@
       font-size: 1.1rem;
     }
 
-    input[type="file"], textarea {
+    input[type="file"],
+    textarea {
       width: 100%;
       padding: 14px;
       border-radius: 14px;
       border: 1px solid var(--border);
       font-size: 0.95rem;
-      background: #fafafa;
+      background: var(--card);
+      color: var(--text);
     }
 
     textarea {
@@ -104,7 +124,9 @@
       transition: all .2s ease;
     }
 
-    button.secondary { background: var(--secondary); }
+    button.secondary {
+      background: var(--secondary);
+    }
 
     button:disabled {
       opacity: .5;
@@ -113,7 +135,7 @@
 
     button:hover:not(:disabled) {
       transform: translateY(-1px);
-      box-shadow: 0 4px 12px rgba(79,70,229,.2);
+      box-shadow: 0 4px 12px rgba(79, 70, 229, .2);
     }
 
     .status {
@@ -126,10 +148,11 @@
       margin-top: 24px;
       padding: 22px;
       border-radius: 16px;
-      background: #fff;
+      background: var(--card);
       border: 1px solid var(--border);
       line-height: 1.6;
       box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.05);
+      color: var(--text);
     }
 
     .confidence-badge {
@@ -163,120 +186,172 @@
     }
 
     @keyframes pulse {
-      0% { opacity: .4 }
-      50% { opacity: 1 }
-      100% { opacity: .4 }
+      0% {
+        opacity: .4
+      }
+
+      50% {
+        opacity: 1
+      }
+
+      100% {
+        opacity: .4
+      }
+    }
+
+    .theme-toggle {
+      position: fixed;
+      top: 20px;
+      right: 20px;
+      background: var(--card);
+      border: 1px solid var(--border);
+      border-radius: 12px;
+      padding: 10px;
+      cursor: pointer;
+      font-size: 1.4rem;
+      transition: transform 0.2s ease;
+      box-shadow: 0 4px 12px rgba(0, 0, 0, 0.1);
+    }
+
+    .theme-toggle:hover {
+      transform: scale(1.1);
     }
   </style>
 </head>
 
 <body>
-<div class="container">
-  <h1>Gemini RAG Assistant</h1>
-  <div class="subtitle">Upload documents · Ask questions · Get grounded answers</div>
-
-  <div class="card">
-    <h3>1. Upload Knowledge</h3>
-    <input type="file" id="files" multiple accept=".pdf,.txt"/>
-    <div class="row">
-      <button id="uploadBtn" onclick="upload()">Upload & Index Files</button>
+  <button class="theme-toggle" onclick="toggleTheme()" title="Toggle dark mode">🌙</button>
+  <div class="container">
+    <h1>Gemini RAG Assistant</h1>
+    <div class="subtitle">Upload documents · Ask questions · Get grounded answers · <a href="/frontend/analytics.html"
+        style="color: var(--primary); text-decoration: none; font-weight: 600;">📊 Analytics</a></div>
+
+    <div class="card">
+      <h3>1. Upload Knowledge</h3>
+      <input type="file" id="files" multiple accept=".pdf,.txt" />
+      <div class="row">
+        <button id="uploadBtn" onclick="upload()">Upload & Index Files</button>
+      </div>
+      <div id="uploadStatus" class="status"></div>
     </div>
-    <div id="uploadStatus" class="status"></div>
-  </div>
 
-  <div class="card">
-    <h3>2. Ask or Summarize</h3>
-    <textarea id="question" placeholder="E.g., 'What are the main risks?' or 'Summarize the document'"></textarea>
-    <div class="row">
-      <button id="askBtn" onclick="ask()">Ask Question</button>
-      <button class="secondary" id="sumBtn" onclick="summarize()">Generate Summary</button>
+    <div class="card">
+      <h3>2. Ask or Summarize</h3>
+      <textarea id="question" placeholder="E.g., 'What are the main risks?' or 'Summarize the document'"></textarea>
+      <div class="row">
+        <button id="askBtn" onclick="ask()">Ask Question</button>
+        <button class="secondary" id="sumBtn" onclick="summarize()">Generate Summary</button>
+      </div>
     </div>
-  </div>
 
-  <div id="answerBox" class="answer" style="display:none;"></div>
+    <div id="answerBox" class="answer" style="display:none;"></div>
 
-</div>
+  </div>
+
+  <script>
+    // ===== THEME TOGGLE =====
+    function toggleTheme() {
+      const html = document.documentElement;
+      const currentTheme = html.getAttribute('data-theme');
+      const newTheme = currentTheme === 'dark' ? 'light' : 'dark';
 
-<script>
-  let busy = false;
+      html.setAttribute('data-theme', newTheme);
+      localStorage.setItem('theme', newTheme);
 
-  function setBusy(state) {
-    busy = state;
-    document.getElementById("askBtn").disabled = state;
-    document.getElementById("sumBtn").disabled = state;
-    document.getElementById("uploadBtn").disabled = state;
-  }
+      // Update button icon
+      const btn = document.querySelector('.theme-toggle');
+      btn.textContent = newTheme === 'dark' ? '☀️' : '🌙';
+    }
 
-  async function upload() {
-    const files = document.getElementById("files").files;
-    if (!files.length) {
-      alert("Please select files first.");
-      return;
+    // Load saved theme on page load
+    (function () {
+      const savedTheme = localStorage.getItem('theme') || 'light';
+      document.documentElement.setAttribute('data-theme', savedTheme);
+      const btn = document.querySelector('.theme-toggle');
+      if (btn) btn.textContent = savedTheme === 'dark' ? '☀️' : '🌙';
+    })();
+
+    // ===== APP LOGIC =====
+    let busy = false;
+
+    function setBusy(state) {
+      busy = state;
+      document.getElementById("askBtn").disabled = state;
+      document.getElementById("sumBtn").disabled = state;
+      document.getElementById("uploadBtn").disabled = state;
     }
 
-    setBusy(true);
-    const statusDiv = document.getElementById("uploadStatus");
-    statusDiv.innerText = "Indexing documents... this may take a moment.";
+    async function upload() {
+      const files = document.getElementById("files").files;
+      if (!files.length) {
+        alert("Please select files first.");
+        return;
+      }
+
+      setBusy(true);
+      const statusDiv = document.getElementById("uploadStatus");
+      statusDiv.innerText = "Indexing documents... this may take a moment.";
 
-    const fd = new FormData();
-    for (let f of files) fd.append("files", f);
+      const fd = new FormData();
+      for (let f of files) fd.append("files", f);
 
-    try {
-      const res = await fetch("/upload", { method: "POST", body: fd });
-      if (!res.ok) throw new Error("Upload failed");
-      const data = await res.json();
-      statusDiv.innerText = data.message || "Done ✅";
-    } catch (e) {
-      statusDiv.innerText = "Error uploading files.";
+      try {
+        const res = await fetch("/upload", { method: "POST", body: fd });
+        if (!res.ok) throw new Error("Upload failed");
+        const data = await res.json();
+        statusDiv.innerText = data.message || "Done ✅";
+      } catch (e) {
+        statusDiv.innerText = "Error uploading files.";
+      }
+      setBusy(false);
     }
-    setBusy(false);
-  }
 
-  async function ask() {
-    const q = document.getElementById("question").value.trim();
-    if (!q) return;
+    async function ask() {
+      const q = document.getElementById("question").value.trim();
+      if (!q) return;
 
-    setBusy(true);
-    const box = document.getElementById("answerBox");
-    box.style.display = "block";
-    box.innerHTML = "<span class='loader'>Thinking...</span>";
+      setBusy(true);
+      const box = document.getElementById("answerBox");
+      box.style.display = "block";
+      box.innerHTML = "<span class='loader'>Thinking...</span>";
 
-    try {
-      const res = await fetch("/ask", {
-        method: "POST",
-        headers: { "Content-Type": "application/json" },
-        body: JSON.stringify({ prompt: q })
-      });
+      try {
+        const res = await fetch("/ask", {
+          method: "POST",
+          headers: { "Content-Type": "application/json" },
+          body: JSON.stringify({ prompt: q })
+        });
 
-      const data = await res.json();
+        const data = await res.json();
 
-      let html = `<div><strong>Answer:</strong><br>${data.answer.replace(/\n/g, '<br>')}</div>`;
+        let html = `<div><strong>Answer:</strong><br>${data.answer.replace(/\n/g, '<br>')}</div>`;
 
-      if (data.confidence > 0) {
-        html += `<div class="confidence-badge">Confidence: ${(data.confidence * 100).toFixed(0)}%</div>`;
-      }
+        if (data.confidence > 0) {
+          html += `<div class="confidence-badge">Confidence: ${(data.confidence * 100).toFixed(0)}%</div>`;
+        }
 
-      if (data.citations && data.citations.length > 0) {
-        html += `<div class="citations"><strong>Sources:</strong><ul>`;
-        data.citations.forEach(c => {
-          html += `<li>${c.source} (Page ${c.page})</li>`;
-        });
-        html += `</ul></div>`;
-      }
+        if (data.citations && data.citations.length > 0) {
+          html += `<div class="citations"><strong>Sources:</strong><ul>`;
+          data.citations.forEach(c => {
+            html += `<li>${c.source} (Page ${c.page})</li>`;
+          });
+          html += `</ul></div>`;
+        }
 
-      box.innerHTML = html;
+        box.innerHTML = html;
 
-    } catch (e) {
-      box.innerText = "⚠️ Error communicating with the server.";
-    }
+      } catch (e) {
+        box.innerText = "⚠️ Error communicating with the server.";
+      }
 
-    setBusy(false);
-  }
+      setBusy(false);
+    }
 
-  function summarize() {
-    document.getElementById("question").value = "Summarize the uploaded documents";
-    ask();
-  }
-</script>
+    function summarize() {
+      document.getElementById("question").value = "Summarize the uploaded documents";
+      ask();
+    }
+  </script>
 </body>
+
 </html>

main.py

@@ -9,6 +9,8 @@ from dotenv import load_dotenv
 import google.generativeai as genai
 
 from rag_store import ingest_documents, search_knowledge, get_all_chunks, clear_database
+from eval_logger import log_eval
+from analytics import get_analytics
 
 # =========================================================
 # ENV + MODEL SETUP
@@ -19,6 +21,11 @@ genai.configure(api_key=os.getenv("GEMINI_API_KEY"))
 MODEL_NAME = "gemini-2.5-flash"
 USE_MOCK = False # Set to False to use real API
 
+# =========================================================
+# FILE UPLOAD LIMITS
+# =========================================================
+MAX_FILE_SIZE = 50 * 1024 * 1024  # 50 MB
+
 # =========================================================
 # APP
 # =========================================================
@@ -53,12 +60,17 @@ def serve_ui():
     with open("frontend/index.html", "r", encoding="utf-8") as f:
         return f.read()
 
+@app.get("/analytics")
+def analytics():
+    """Return analytics data from evaluation logs."""
+    return get_analytics()
+
 # ---------------------------------------------------------
 # UPLOAD
 # ---------------------------------------------------------
 @app.post("/upload")
 async def upload(files: list[UploadFile] = File(...)):
-    # 1. VALIDATION: Strict File Type Check
+    # 1. VALIDATION: File Type and Size Check
     for file in files:
         ext = file.filename.split(".")[-1].lower()
         if ext not in ["pdf", "txt"]:
@@ -66,6 +78,19 @@ async def upload(files: list[UploadFile] = File(...)):
                 status_code=400, 
                 content={"error": f"Invalid file type: '{file.filename}'. Only .pdf and .txt files are allowed."}
             )
+        
+        # Check file size
+        file.file.seek(0, 2)  # Seek to end
+        file_size = file.file.tell()
+        file.file.seek(0)  # Reset to beginning
+        
+        if file_size > MAX_FILE_SIZE:
+            size_mb = file_size / (1024 * 1024)
+            max_mb = MAX_FILE_SIZE / (1024 * 1024)
+            return JSONResponse(
+                status_code=413,
+                content={"error": f"File '{file.filename}' is too large ({size_mb:.1f} MB). Maximum size is {max_mb:.0f} MB."}
+            )
 
     try:
         # 2. CLEAR CONTEXT: Start fresh for every upload session
@@ -126,13 +151,31 @@ async def ask(data: PromptRequest):
                 return response
             except Exception as e:
                 err_str = str(e)
+                
+                # API Key Issues
+                if "API_KEY" in err_str or "invalid" in err_str.lower() and "key" in err_str.lower():
+                    raise ValueError("Invalid API key. Please check your GEMINI_API_KEY in the .env file.")
+                
+                # Quota Exhausted
+                if "quota" in err_str.lower() or "limit" in err_str.lower():
+                    raise ValueError("API quota exhausted. Please try again later or upgrade your API plan.")
+                
+                # Rate Limiting (429)
                 if "429" in err_str:
                     if attempt < retries:
                         wait_time = base_delay * (2 ** attempt)
                         print(f"DEBUG: 429 Rate limit hit. Retrying in {wait_time}s...")
                         pytime.sleep(wait_time)
                         continue
-                raise e
+                    else:
+                        raise ValueError("Rate limit exceeded. Please try again in a few minutes.")
+                
+                # Safety Filters
+                if "safety" in err_str.lower() or "blocked" in err_str.lower():
+                    raise ValueError("Content was blocked by safety filters. Please rephrase your question.")
+                
+                # Generic error
+                raise ValueError(f"LLM API error: {err_str}")
 
     if is_summary:
         chunks = get_all_chunks(limit=80)
@@ -180,10 +223,18 @@ Content:
             answer_cache[key] = (now, response)
             return response
             
-        except Exception as e:
+        except ValueError as e:
+            # User-friendly error from generate_safe
             print(f"Summary failed: {e}")
             return JSONResponse(status_code=200, content={
-                "answer": f"System is currently overloaded (Rate Limit). Please try again in a minute.\nDetails: {str(e)}",
+                "answer": str(e),
+                "confidence": 0.0,
+                "citations": []
+            })
+        except Exception as e:
+            print(f"Summary failed: {e}")
+            return JSONResponse(status_code=500, content={
+                "answer": f"An unexpected error occurred: {str(e)}",
                 "confidence": 0.0,
                 "citations": []
             })
@@ -199,6 +250,14 @@ Content:
             "confidence": 0.0,
             "citations": []
         }
+
+        log_eval(
+            query=prompt_text,
+            retrieved_count=0,
+            confidence=0.0,
+            answer_known=False
+        )
+
         answer_cache[key] = (now, response)
         return response
 
@@ -219,8 +278,29 @@ Context:
 Question:
 {prompt_text}
 """
-    llm = model.generate_content(prompt)
-    answer_text = llm.text
+    llm = None
+    answer_text = ""
+    
+    try:
+        llm = model.generate_content(prompt)
+        answer_text = llm.text
+    except ValueError as e:
+        # User-friendly error from API
+        response = {
+            "answer": str(e),
+            "confidence": 0.0,
+            "citations": []
+        }
+        answer_cache[key] = (now, response)
+        return response
+    except Exception as e:
+        # Unexpected error
+        response = {
+            "answer": f"An unexpected error occurred: {str(e)}",
+            "confidence": 0.0,
+            "citations": []
+        }
+        return JSONResponse(status_code=500, content=response)
     
     # Fix Fake Confidence: If the model says "I don't know", confidence should be 0.
     confidence = round(min(1.0, len(results) / 5), 2)
@@ -236,5 +316,14 @@ Question:
         }.values())
     }
 
+    answer_known = "i don't know" not in answer_text.lower()
+
+    log_eval(
+        query=prompt_text,
+        retrieved_count=len(results),
+        confidence=confidence,
+        answer_known=answer_known
+    )
+
     answer_cache[key] = (now, response)
     return response

rag_eval_logs.jsonl

@@ -0,0 +1,23 @@
+{"timestamp": 1767711001.7256067, "query": "how much did student got the scholarship?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767711040.2298725, "query": "what is program duration and fees?", "retrieved_count": 5, "confidence": 0.0, "answer_known": false}
+{"timestamp": 1767711058.4562092, "query": "what is program duration?", "retrieved_count": 5, "confidence": 0.0, "answer_known": false}
+{"timestamp": 1767711112.7273183, "query": "what is university name?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767711173.2461476, "query": "what is university name and tution fees?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767711339.3448784, "query": "what is the scholarship  name?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767711360.8415213, "query": "how much for OSHC?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767711421.3609428, "query": "what are the conditions to accept the offer?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767711528.7769852, "query": "is this Sydney university?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767711553.6641674, "query": "what is  the father name?", "retrieved_count": 1, "confidence": 0.0, "answer_known": false}
+{"timestamp": 1767711646.590812, "query": "tell me about refund process?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767711707.6674147, "query": "when is orientation and enrollment?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767711808.9429202, "query": "is it good university?", "retrieved_count": 5, "confidence": 0.0, "answer_known": false}
+{"timestamp": 1767714229.559012, "query": "what is candidate name and college name?", "retrieved_count": 1, "confidence": 0.0, "answer_known": false}
+{"timestamp": 1767774219.903037, "query": "what is university name?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767774233.0324135, "query": "what are my risks?", "retrieved_count": 0, "confidence": 0.0, "answer_known": false}
+{"timestamp": 1767774272.1751444, "query": "what is course name?and where is monash university?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767775699.7483144, "query": "Summarize the uploaded documents", "retrieved_count": 0, "confidence": 0.0, "answer_known": false}
+{"timestamp": 1767775793.0641322, "query": "Summarize the uploaded documents", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767775929.047869, "query": "what is university name?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767776180.9555495, "query": "what are the visa conditions?", "retrieved_count": 5, "confidence": 0.0, "answer_known": false}
+{"timestamp": 1767776250.0441537, "query": "tell me about program information?", "retrieved_count": 5, "confidence": 1.0, "answer_known": true}
+{"timestamp": 1767777566.4153016, "query": "what was the role ?", "retrieved_count": 3, "confidence": 0.6, "answer_known": true}

rag_store.py

@@ -1,19 +1,14 @@
 import faiss
-import numpy as np
 import os
 import pickle
 from pypdf import PdfReader
-from sentence_transformers import SentenceTransformer
+from sentence_transformers import SentenceTransformer, CrossEncoder
 
-# -----------------------
-# Global state
-# -----------------------
-index = None
-documents = []
-metadata = []
-
-# Using a lightweight, high-performance embedding model
-embedder = SentenceTransformer("all-MiniLM-L6-v2")
+# =========================================================
+# CONFIG
+# =========================================================
+USE_HNSW = True
+USE_RERANKER = True
 
 CHUNK_SIZE = 800
 CHUNK_OVERLAP = 200
@@ -21,166 +16,177 @@ CHUNK_OVERLAP = 200
 DB_FILE_INDEX = "vector.index"
 DB_FILE_META = "metadata.pkl"
 
-# -----------------------
-# Helpers
-# -----------------------
+# =========================================================
+# GLOBAL STATE
+# =========================================================
+index = None
+documents = []
+metadata = []
+
+embedder = SentenceTransformer("all-MiniLM-L6-v2")
+reranker = CrossEncoder("cross-encoder/ms-marco-MiniLM-L-6-v2")
+
+# =========================================================
+# HELPERS
+# =========================================================
 def chunk_text(text):
-    """Splits text into overlapping chunks."""
-    chunks = []
-    start = 0
-    while start < len(text):
-        end = start + CHUNK_SIZE
-        chunks.append(text[start:end])
-        start += CHUNK_SIZE - CHUNK_OVERLAP
+    import re
+    sentences = re.split(r'(?<=[.!?])\s+', text)
+
+    chunks, current = [], ""
+    for s in sentences:
+        if len(current) + len(s) > CHUNK_SIZE and current:
+            chunks.append(current.strip())
+            overlap = max(0, len(current) - CHUNK_OVERLAP)
+            current = current[overlap:] + " " + s
+        else:
+            current += " " + s if current else s
+
+    if current.strip():
+        chunks.append(current.strip())
     return chunks
 
+
 def save_db():
-    global index, documents, metadata
     if index:
         faiss.write_index(index, DB_FILE_INDEX)
     if documents:
         with open(DB_FILE_META, "wb") as f:
             pickle.dump({"documents": documents, "metadata": metadata}, f)
-    print("DEBUG: Knowledge base saved to disk.")
+
 
 def load_db():
     global index, documents, metadata
     if os.path.exists(DB_FILE_INDEX) and os.path.exists(DB_FILE_META):
-        try:
-            index = faiss.read_index(DB_FILE_INDEX)
-            with open(DB_FILE_META, "rb") as f:
-                data = pickle.load(f)
-                documents = data["documents"]
-                metadata = data["metadata"]
-            print(f"DEBUG: Loaded {len(documents)} documents from disk.")
-        except Exception as e:
-            print(f"DEBUG: Failed to load DB: {e}")
-            index = None
-            documents = []
-            metadata = []
-    else:
-        print("DEBUG: No existing DB found. Starting fresh.")
+        index = faiss.read_index(DB_FILE_INDEX)
+        with open(DB_FILE_META, "rb") as f:
+            data = pickle.load(f)
+            documents = data["documents"]
+            metadata = data["metadata"]
+        print(f"DEBUG: Loaded {len(documents)} chunks")
+
 
-# Auto-load on startup
 load_db()
 
+
 def clear_database():
     global index, documents, metadata
     index = None
     documents = []
     metadata = []
-    
-    # Remove persistence files if they exist
+
     if os.path.exists(DB_FILE_INDEX):
         os.remove(DB_FILE_INDEX)
     if os.path.exists(DB_FILE_META):
         os.remove(DB_FILE_META)
-    
-    print("DEBUG: Database cleared.")
 
-# -----------------------
-# Ingest
-# -----------------------
+
+# =========================================================
+# INGEST
+# =========================================================
 def ingest_documents(files):
     global index, documents, metadata
 
-    texts = []
-    meta = []
+    texts, meta = [], []
 
     for file in files:
-        filename = file.filename
-        
-        # Handle PDFs
-        if filename.endswith(".pdf"):
+        name = file.filename
+
+        if name.endswith(".pdf"):
             reader = PdfReader(file.file)
             for i, page in enumerate(reader.pages):
-                page_text = page.extract_text()
-                if page_text:
-                    for chunk in chunk_text(page_text):
+                try:
+                    text = page.extract_text()
+                except Exception:
+                    text = None
+
+                if text:
+                    for chunk in chunk_text(text):
                         texts.append(chunk)
-                        meta.append({"source": filename, "page": i + 1})
-        
-        # Handle Text files
-        elif filename.endswith(".txt"):
-            content = file.file.read().decode("utf-8")
+                        meta.append({"source": name, "page": i + 1})
+
+        elif name.endswith(".txt"):
+            content = file.file.read().decode("utf-8", errors="ignore")
             for chunk in chunk_text(content):
                 texts.append(chunk)
-                meta.append({"source": filename, "page": "N/A"})
+                meta.append({"source": name, "page": "N/A"})
 
-    # Check for empty or unreadable content
-    total_length = sum(len(t) for t in texts)
-    if total_length < 50:
+    if not texts:
         raise ValueError(
-            "Extracted text is too short or empty. "
-            "If this is a PDF, it might be a scanned image without a text layer. "
-            "Please use a text-selectable PDF or a .txt file."
+            "No readable text found. "
+            "If this is a scanned PDF, OCR is required."
         )
 
-    if not texts:
-        raise ValueError("No readable text found in documents.")
-
-    # Create Embeddings (Normalized for better cosine similarity)
-    # append to existing if needed, but for now simplistic re-build or append?
-    # Simpler to just ADD to the existing index.
-    
-    new_embeddings = embedder.encode(texts, convert_to_numpy=True, normalize_embeddings=True)
+    embeddings = embedder.encode(
+        texts,
+        convert_to_numpy=True,
+        normalize_embeddings=True
+    )
 
     if index is None:
-        # USE INNER PRODUCT (Cosine Similarity) for normalized vectors
-        index = faiss.IndexFlatIP(new_embeddings.shape[1])
-    
-    index.add(new_embeddings)
-
+        dim = embeddings.shape[1]
+        if USE_HNSW:
+            index = faiss.IndexHNSWFlat(dim, 32)
+            index.hnsw.efConstruction = 200
+            index.hnsw.efSearch = 64
+        else:
+            index = faiss.IndexFlatIP(dim)
+
+    index.add(embeddings)
     documents.extend(texts)
     metadata.extend(meta)
 
     save_db()
-    
     return len(documents)
 
-# -----------------------
-# Q&A Search (filtered)
-# -----------------------
-def search_knowledge(query, top_k=5, min_similarity=0.3):
+
+# =========================================================
+# SEARCH
+# =========================================================
+def search_knowledge(query, top_k=8, min_similarity=0.25):
     if index is None:
         return []
 
-    # SEARCH with normalized query
-    query_vec = embedder.encode([query], convert_to_numpy=True, normalize_embeddings=True)
-    
-    # FAISS returns scores (dot product), which = cosine similarity for normalized vectors
-    scores, indices = index.search(query_vec, top_k)
+    qvec = embedder.encode(
+        [query],
+        convert_to_numpy=True,
+        normalize_embeddings=True
+    )
+
+    scores, indices = index.search(qvec, top_k)
+    candidates = []
+    ql = query.lower()
 
-    results = []
-    print(f"DEBUG: Query: '{query}'")
     for idx, score in zip(indices[0], scores[0]):
-        if idx == -1: continue # FAISS padding
-        
-        print(f"DEBUG: Found chunk {idx} with score {score:.4f}")
-        
-        # Filter out results that are too irrelevant (score too low)
-        if score > min_similarity:
-            results.append({
-                "text": documents[idx],
-                "metadata": metadata[idx],
-                "score": float(score)
+        if idx == -1:
+            continue
+
+        text = documents[idx]
+        meta = metadata[idx]
+        keyword_hits = sum(w in text.lower() for w in ql.split())
+        hybrid_score = float(score) + (0.05 * keyword_hits)
+
+        if hybrid_score >= min_similarity:
+            candidates.append({
+                "text": text,
+                "metadata": meta,
+                "hybrid_score": hybrid_score
             })
 
-    return results
+    if USE_RERANKER and candidates:
+        pairs = [(query, c["text"]) for c in candidates]
+        scores = reranker.predict(pairs)
+        for c, s in zip(candidates, scores):
+            c["rerank"] = float(s)
+        candidates.sort(key=lambda x: x["rerank"], reverse=True)
+    else:
+        candidates.sort(key=lambda x: x["hybrid_score"], reverse=True)
 
-# -----------------------
-# Summary Retrieval (NO FILTER)
-# -----------------------
-def get_all_chunks(limit=50):
-    if not documents:
-        return []
+    return candidates[:5]
 
-    results = []
-    # Return a sample of chunks for summarization
-    for text, meta in zip(documents[:limit], metadata[:limit]):
-        results.append({
-            "text": text,
-            "metadata": meta
-        })
 
-    return results
+def get_all_chunks(limit=50):
+    return [
+        {"text": t, "metadata": m}
+        for t, m in zip(documents[:limit], metadata[:limit])
+    ]

render.yaml

@@ -1,12 +0,0 @@
-services:
-  - type: web
-    name: gemini-rag-fastapi
-    env: python
-    plan: free
-    buildCommand: |
-      pip install -r requirements.txt
-    startCommand: |
-      uvicorn main:app --host 0.0.0.0 --port 10000
-    envVars:
-      - key: GEMINI_API_KEY
-        sync: false