Adding a Chatbot

Dockerfile

@@ -0,0 +1,24 @@
+# Base image: Ruby with necessary dependencies for Jekyll
+FROM ruby:3.2
+
+# Install dependencies
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    nodejs \
+    && rm -rf /var/lib/apt/lists/*
+
+# Set the working directory inside the container
+WORKDIR /usr/src/app
+
+# Copy Gemfile into the container (necessary for `bundle install`)
+COPY Gemfile ./
+
+# Install bundler and dependencies
+RUN gem install bundler:2.3.26 && bundle install
+
+# Expose port 4000 for Jekyll server
+EXPOSE 4000
+
+# Command to serve the Jekyll site
+CMD ["bundle", "exec", "jekyll", "serve", "--host", "0.0.0.0", "--watch"]
+

_includes/navbar.html

@@ -91,6 +91,11 @@
                         See on GitHub
                     </a>
                 </li>
+                <li class="nav-item">
+                    <a class="btn btn-primary" href="/chat">
+                        Chat
+                    </a>
+                </li>
             </ul>
         </div>
     </nav>

backend/infer_api.py

@@ -0,0 +1,127 @@
+import os
+from fastapi import FastAPI, HTTPException
+from pydantic import BaseModel
+from sentence_transformers import SentenceTransformer
+from pinecone import Pinecone
+from groq import Groq
+from fastapi.middleware.cors import CORSMiddleware
+
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+
+# ----------------- API KEYS -----------------
+# Pinecone
+PINECONE_API_KEY = "enter_key"
+PINECONE_ENV = "us-east-1"
+PINECONE_INDEX_NAME  = "qutip-shrody"
+
+
+# Groq settings
+GROQ_API_KEY = os.getenv("GROQ_API_KEY", "enter_key")
+GROQ_MODEL_NAME =  "llama-3.1-8b-instant" #"llama-3.3-70b-versatile"
+
+app = FastAPI()
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# embedding 
+embedder = SentenceTransformer("intfloat/e5-small-v2")
+
+# init pinecone
+pc = Pinecone(api_key=PINECONE_API_KEY, environment=PINECONE_ENV)
+index = pc.Index(PINECONE_INDEX_NAME)
+
+# init groq
+groq_client = Groq(api_key=GROQ_API_KEY)
+
+class ChatRequest(BaseModel):
+    message: str
+
+def embed_text(text: str) -> list[float]:
+    """
+    Returns a normalized embedding for the given text.
+    """
+    embedding = embedder.encode(text, normalize_embeddings=True)
+    return embedding.tolist()
+
+def retrieve_relevant_docs(query_vector: list[float], top_k: int = 5) -> list[str]:
+    """
+    Performs a Pinecone vector search. Returns a list of chunk_text strings
+    (as stored under metadata["chunk_text"]) for the top_k matches.
+    """
+    response = index.query(
+        namespace="__default__",
+        vector=query_vector,
+        top_k=top_k,
+        include_metadata=True
+    )
+
+    docs = []
+    for match in response.get("matches", []):
+        metadata = match.get("metadata", {})
+        if "chunk_text" in metadata:
+            docs.append(metadata["chunk_text"])
+    return docs
+
+@app.post("/api/chat")
+async def chat_endpoint(request: ChatRequest):
+    user_question = request.message.strip()
+    if not user_question:
+        raise HTTPException(status_code=400, detail="Empty message")
+
+    try:
+        q_vec = embed_text(user_question)
+        raw_response = index.query(
+            namespace="__default__",
+            vector=q_vec,
+            top_k=5,
+            include_metadata=True
+        )
+        docs = []
+        for match in raw_response.get("matches", []):
+            if match.get("score", 0) > 0.5: # can be tuned further
+                metadata = match.get("metadata", {})
+                if "chunk_text" in metadata:
+                    docs.append(metadata["chunk_text"])
+
+        # 4) Combine all chunks into a single “context” block
+        context = "\n".join(docs) if docs else ""
+
+        print("Filtered Context:\n", context)
+
+        # 5) Build messages: context goes in system message only
+        system_message = {
+            "role": "system",
+            "content": (
+                "You are Shrody, an expert assistant for QuTiP's (its an open-source python library for simulating the dynamics of open quantum systems) documentation. "
+                "Be friendly, clear, and use bit of emoji. Use only the following context to answer the user's question:\n\n"
+                f"{context}\n\n"
+                "Always be concise and accurate. If the context does not contain the answer, kindly say so.  Don't guess or invent information."
+            )
+        }
+
+        user_message = {
+            "role": "user",
+            "content": user_question
+        }
+
+        # 6) Call Groq’s chat endpoint
+        chat_completion = groq_client.chat.completions.create(
+            messages=[system_message, user_message],
+            model=GROQ_MODEL_NAME,
+        )
+
+        generated_answer = chat_completion.choices[0].message.content
+        return {"response": generated_answer}
+
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+if __name__ == "__main__":
+    import uvicorn
+    # Make sure the module name matches your filename (e.g. if this file is infer_api.py):
+    uvicorn.run("infer_api:app", port=5001, reload=True)

backend/pc_upcert.py

@@ -0,0 +1,98 @@
+import os
+import uuid
+from collections import defaultdict
+
+import pandas as pd
+from unstructured.partition.pdf import partition_pdf
+from unstructured.documents.elements import Table
+from sentence_transformers import SentenceTransformer
+from pinecone import Pinecone
+from tqdm import tqdm
+
+# ----------------- API KEYS -----------------
+# Pinecone
+PINECONE_API_KEY = "enter-key"
+PINECONE_ENV = "us-east-1"
+PINECONE_INDEX_NAME  = "qutip-shrody"
+
+
+# 1. Extract one chunk per PDF page, skip first 10 pages, include tables if present
+def extract_chunks_by_page(pdf_path, skip_pages=10):
+    elements = partition_pdf(filename=pdf_path)
+    pages = defaultdict(list)
+    tables = defaultdict(list)
+
+    # Group text & tables by page number
+    for el in elements:
+        page_no = getattr(el.metadata, "page_number", None)
+        if page_no is None or page_no <= skip_pages:
+            continue
+
+        if isinstance(el, Table):
+            try:
+                df = el.to_dataframe()
+                md = df.to_markdown(index=False)
+                tables[page_no].append(md)
+            except Exception:
+                tables[page_no].append(el.text.strip())
+        elif el.category not in ("PageBreak", "Image"):
+            text = el.text.strip()
+            if text:
+                pages[page_no].append(text)
+
+    # Build one chunk per page
+    chunks = []
+    for page_no in sorted(set(pages) | set(tables)):
+        parts = []
+        if pages.get(page_no):
+            parts.append("\n\n".join(pages[page_no]))
+        if tables.get(page_no):
+            for tbl_md in tables[page_no]:
+                parts.append("\n\n**Table:**\n\n" + tbl_md)
+        joined = "\n\n".join(parts).strip()
+        if joined:
+            chunks.append({
+                "text": joined,
+                "page_number": page_no
+            })
+    return chunks
+
+# 2. Load embedding model
+model = SentenceTransformer("intfloat/e5-small-v2")
+def get_embedding(text):
+    return model.encode([text], normalize_embeddings=True)[0]
+
+# 3. Initialize Pinecone (replace with your key & environment)
+pc = Pinecone(api_key=PINECONE_API_KEY, environment=PINECONE_ENV)
+index = pc.Index(PINECONE_INDEX_NAME)
+
+# 4. Embed and upsert into Pinecone
+def upsert_pdf_by_page(pdf_path, base_doc_id="mydoc"):
+    chunks = extract_chunks_by_page(pdf_path, skip_pages=10)
+    print(f"🧩 Extracted {len(chunks)} page-level chunks (skipped first 10 pages).")
+
+    vectors = []
+    for chunk in tqdm(chunks, desc="Embedding pages"):
+        chunk_id = f"{base_doc_id}-page{chunk['page_number']}"
+        emb = get_embedding(chunk["text"])
+        vectors.append({
+            "id": chunk_id,
+            "values": emb,
+            "metadata": {
+                "source": base_doc_id,
+                "page_number": chunk["page_number"],
+                "chunk_text": chunk["text"]
+            }
+        })
+
+    # Batch upserts
+    batch_size = 50
+    for start in range(0, len(vectors), batch_size):
+        batch = vectors[start:start+batch_size]
+        index.upsert(vectors=batch)
+        print(f"✅ Upserted pages {start}–{start + len(batch) - 1}")
+
+if __name__ == "__main__":
+    pdf_file = "qutip.pdf"
+    doc_id = "qutip-5.1"
+    upsert_pdf_by_page(pdf_file, base_doc_id=doc_id)

backend/requirements.txt

@@ -0,0 +1,9 @@
+fastapi==0.115.12
+groq==0.26.0
+pandas==2.3.0
+pinecone==7.0.2
+pydantic==2.11.5
+sentence_transformers==4.1.0
+tqdm==4.67.0
+unstructured[full-docs]==0.17.2
+uvicorn==0.34.3

chat.md

@@ -0,0 +1,86 @@
+---
+layout: default
+title: Chat with Docs
+permalink: /chat/
+---
+
+<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.7.0/styles/default.min.css">
+<script src="https://cdnjs.cloudflare.com/ajax/libs/marked/4.3.0/marked.min.js"></script>
+<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.7.0/highlight.min.js"></script>
+<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.7.0/languages/python.min.js"></script>
+
+<div id="chat-container">
+  <h2 class="mb-4">📚 QuTiP's Schrödy</h2>
+  <div id="messages" class="mb-3"></div>
+
+  <div id="input-group" class="input-group">
+    <input
+      type="text"
+      id="userInput"
+      class="form-control"
+      placeholder="Type your question..."
+      onkeypress="if(event.key === 'Enter') sendMessage()"
+    />
+    <button class="btn btn-primary" onclick="sendMessage()">Send</button>
+  </div>
+</div>
+
+<script>
+  hljs.highlightAll();
+
+  function appendMessage(role, markdownText) {
+    const messagesDiv = document.getElementById("messages");
+    const html = marked.parse(markdownText);
+
+    const wrapper = document.createElement("div");
+    wrapper.classList.add("message", role);
+
+    const bubble = document.createElement("div");
+    bubble.classList.add("bubble");
+    bubble.innerHTML = html;
+
+    wrapper.appendChild(bubble);
+    messagesDiv.appendChild(wrapper);
+
+    messagesDiv.scrollTop = messagesDiv.scrollHeight;
+    hljs.highlightAll(); // Re-highlight after adding new content
+  }
+
+  async function sendMessage() {
+    const input = document.getElementById("userInput");
+    const userText = input.value.trim();
+    if (!userText) return;
+
+    appendMessage("user", userText);
+    input.value = "";
+    input.disabled = true;
+
+    try {
+      const resp = await fetch("http://localhost:5001/api/chat", {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({ message: userText }),
+      });
+
+      if (!resp.ok) throw new Error(`HTTP ${resp.status}`);
+      const data = await resp.json();
+
+      appendMessage("bot", data.response);
+    } catch (err) {
+      console.error("Error calling /api/chat:", err);
+      appendMessage("bot", "❗ Sorry, something went wrong. Please try again later.");
+    } finally {
+      input.disabled = false;
+      input.focus();
+    }
+  }
+
+  window.addEventListener("DOMContentLoaded", () => {
+    appendMessage(
+      "bot",
+      `👋 Hi, I'm **Shrody**, QuTiP's doc assistant!  
+        I'm here to help you understand and explore the QuTiP documentation.  
+        Just ask me a question about how things work—functions, features, usage examples—and I'll do my best to guide you!`
+    );
+  });
+</script>