Update app.py

app.py

@@ -1,22 +1,22 @@
-# app.py  ──────────────────────────────────────────────────────────────
-# Requires: gradio<=3.31.0  (later versions strip target="_blank")
-#           architecture.png  (diagram you supply)  ─ place beside app.py
+# app.py ────────────────────────────────────────────────────────────────
+# Pin Gradio ≤ 3.31.0 in requirements.txt so <a target="_blank"> is kept
+# and place architecture.png beside this file.
 import os, re, faiss, zipfile, warnings, gradio as gr
 from pathlib import Path
-from typing  import List
+from typing import List
 from sentence_transformers import SentenceTransformer
-from PyPDF2   import PdfReader
-from docx     import Document
+from PyPDF2 import PdfReader
+from docx import Document
 from docx.opc.exceptions import PackageNotFoundError
-from openai   import OpenAI
+from openai import OpenAI
 
-# ───────── 0. résumé → text ───────────────────────────────────────────
-FILE = Path("my_resume.pdf")                       # already in repo
+# ───────── 0. résumé → plain-text ──────────────────────────────────────
+FILE = Path("my_resume.pdf")
 
-def read_pdf(p: Path)->str:
+def read_pdf(p: Path) -> str:
     return " ".join(pg.extract_text() or "" for pg in PdfReader(p).pages)
 
-def read_docx(p: Path)->str:
+def read_docx(p: Path) -> str:
     return " ".join(par.text for par in Document(p).paragraphs if par.text.strip())
 
 try:
@@ -27,84 +27,102 @@ except (PackageNotFoundError, KeyError, zipfile.BadZipFile):
 
 text = re.sub(r"\s+", " ", raw).strip()
 
-# Extra searchable metadata (Markdown links open in a new tab with ctrl/cmd-click)
-LINKEDIN = "[LinkedIn Profile](https://www.linkedin.com/in/sriharideep/)"
-BLOG     = "[Technical Blog](https://sfdcbrewery.github.io/)"
-ARCH_MD  = "![Architecture Diagram](architecture.png)"
-
+# ───────── 0-bis. extra searchable metadata ───────────────────────────
+LINK_MD = '<a href="https://www.linkedin.com/in/sriharideep/" target="_blank">' \
+          'LinkedIn Profile</a>'
+BLOG_MD = '<a href="https://sfdcbrewery.github.io/" target="_blank">' \
+          'Technical Blog</a>'
 ARCH_NOTE = (
     "ARCHITECTURE NOTE – The bot follows a Retrieval-Augmented Generation "
     "(RAG) design: PDF → 180-token chunks → MiniLM-L6 embeddings → FAISS "
-    "similarity search → GPT-3.5-turbo answer constrained to retrieved context."
+    "similarity search → GPT-3.5-turbo answer constrained to context."
 )
+ARCH_MD = f"{ARCH_NOTE}\n\n![Architecture Diagram](architecture.png)"
 
-# Append so they are embedded and retrievable
-text += f" LinkedIn: {LINKEDIN} Blog: {BLOG} {ARCH_NOTE} {ARCH_MD}"
+# make them retrievable by the RAG index (even though we’ll short-circuit)
+text += f" LinkedIn: {LINK_MD} Blog: {BLOG_MD} {ARCH_MD}"
 
-# ───────── 1. chunk → FAISS ───────────────────────────────────────────
-def chunkify(t: str, max_tok: int = 180)->List[str]:
+# ───────── 1. text → embeddings → FAISS ───────────────────────────────
+def chunkify(t: str, max_tok: int = 180) -> List[str]:
     out, buf, n = [], [], 0
     for s in re.split(r"(?<=[.!?])\s+", t):
         w = len(s.split())
         if n + w > max_tok:
             out.append(" ".join(buf)); buf, n = [], 0
         buf.append(s); n += w
-    if buf: out.append(" ".join(buf))
+    if buf:
+        out.append(" ".join(buf))
     return out
 
 CHUNKS = chunkify(text)
 
-embed = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
-vecs  = embed.encode(CHUNKS, convert_to_numpy=True)
+embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
+vecs = embedder.encode(CHUNKS, convert_to_numpy=True)
 faiss.normalize_L2(vecs)
 index = faiss.IndexFlatIP(vecs.shape[1]); index.add(vecs)
 
-def retrieve(q: str, k:int=4):
-    qv = embed.encode([q], convert_to_numpy=True); faiss.normalize_L2(qv)
+def retrieve(q: str, k: int = 4):
+    qv = embedder.encode([q], convert_to_numpy=True); faiss.normalize_L2(qv)
     sims, idx = index.search(qv, k)
     return sims[0], [CHUNKS[i] for i in idx[0]]
 
-# ───────── 2. OpenAI client ───────────────────────────────────────────
+# ───────── 2. OpenAI client ────────────────────────────────────────────
 client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
 MODEL  = "gpt-3.5-turbo-0125"
-
 SYSTEM = ("You are a helpful assistant. Answer ONLY with facts in the context. "
           "If missing, reply exactly: \"I don't know based on the resume.\"")
 
-def overlap(a: str, b: str)->bool:
+def overlap(a: str, b: str) -> bool:
     return bool(set(re.findall(r"\w+", a.lower())) &
                 set(re.findall(r"\w+", b.lower())))
 
+# ───────── 2-bis. guard words & static answers ─────────────────────────
 SAFE = {"experience","project","certification","certifications","education",
         "skill","skills","summary","company","companies","role","linkedin",
         "website","blog","portfolio","architecture"}
 
-def generate(msg: str)->str:
-    # 0. Must be résumé-related
-    if not (SAFE & set(re.findall(r"\w+", msg.lower()))):
+STATIC_ANSWERS = {
+    "linkedin": LINK_MD,
+    "linked-in": LINK_MD,
+    "blog": BLOG_MD,
+    "architecture": ARCH_MD
+}
+
+# ───────── 2-ter. generator ───────────────────────────────────────────
+def generate(msg: str) -> str:
+    lower_msg = msg.lower().strip()
+
+    # A. serve static responses verbatim
+    for key, val in STATIC_ANSWERS.items():
+        if key in lower_msg:
+            return val
+
+    # B. resume-related check
+    if not (SAFE & set(re.findall(r"\w+", lower_msg))):
         return "Please ask something related to my résumé."
 
-    # 1. Retrieve
+    # C. retrieve
     sims, ctxs = retrieve(msg)
-    thresh = 0.10 if len(msg.split()) < 3 else 0.25
-    if max(sims) < thresh:
+    min_sim = 0.10 if len(msg.split()) < 3 else 0.25
+    if max(sims) < min_sim:
         return "I don't know based on the resume."
 
-    # 2. LLM
-    ctx  = "\n".join(ctxs)
-    ans  = client.chat.completions.create(
+    # D. GPT-3.5-turbo
+    ctx = "\n".join(ctxs)
+    ans = client.chat.completions.create(
         model=MODEL,
         messages=[
-            {"role":"system","content":SYSTEM},
-            {"role":"user", "content":f"Context:\n{ctx}"},
-            {"role":"user", "content":f"Question: {msg}"}
+            {"role": "system", "content": SYSTEM},
+            {"role": "user",   "content": f"Context:\n{ctx}"},
+            {"role": "user",   "content": f"Question: {msg}"}
         ],
-        max_tokens=256, temperature=0.2
+        max_tokens=256,
+        temperature=0.2
     ).choices[0].message.content.strip()
 
     return ans if overlap(ans, ctx) else "I don't know based on the resume."
 
-# ───────── 3. Gradio UI ───────────────────────────────────────────────
+# ───────── 3. Gradio UI ────────────────────────────────────────────────
 quick = [
     "Professional Summary","Education details","Experience",
     "Certifications","Skills","LinkedIn","Blog","Architecture"
@@ -121,7 +139,7 @@ with gr.Blocks(theme="soft") as demo:
             inp   = gr.Textbox(placeholder="Ask about my résumé…", show_label=False)
             state = gr.State([])
 
-    # ENTER key
+    # ENTER
     def user_submit(msg, hist):
         ans  = generate(msg)
         hist = hist + [{"role":"user","content":msg},
@@ -132,7 +150,7 @@ with gr.Blocks(theme="soft") as demo:
 
     # QUICK buttons
     def quick_send(hist, q):
-        ans  = generate(q)
+        ans = generate(q)
         hist = hist + [{"role":"user","content":q},
                        {"role":"assistant","content":ans}]
         return hist, hist
@@ -141,4 +159,5 @@ with gr.Blocks(theme="soft") as demo:
         b.click(quick_send, [state, gr.State(q)], [chat, state])
 
 if __name__ == "__main__":
-    demo.launch(share=True)   # remove share=True if not needed
+    # When running in HF Spaces, share=True is ignored; safe to leave as-is.
+    demo.launch(share=True)