Spaces:
No application file
No application file
File size: 11,418 Bytes
8a183f5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 |
import os, io, glob, time, hashlib
import pandas as pd
import numpy as np
from typing import List, Tuple
from dataclasses import dataclass
from dotenv import load_dotenv
# Telegram
from telegram import Update
from telegram.constants import ParseMode
from telegram.ext import Application, CommandHandler, MessageHandler, ContextTypes, filters
# OpenAI
from openai import OpenAI
# Files / parsing
from docx import Document as DocxDocument
from pypdf import PdfReader
# Vector store
import faiss
load_dotenv()
import httpx
if not hasattr(httpx, "proxies"):
httpx.proxies = None
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
TELEGRAM_BOT_TOKEN = os.getenv("TELEGRAM_BOT_TOKEN")
OPENAI_MODEL = os.getenv("OPENAI_MODEL", "gpt-4o-mini")
EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL", "text-embedding-3-small")
STRICT_DOC_MODE = (os.getenv("STRICT_DOC_MODE", "true").lower() == "true")
DOCS_DIR = os.getenv("DOCS_DIR", "wedding_docs")
INDEX_PATH = os.getenv("INDEX_PATH", "wedding.index")
META_CSV = os.getenv("META_CSV", "wedding_chunks.csv")
client = OpenAI(api_key=OPENAI_API_KEY)
# -----------------------------
# Utilities to read documents
# -----------------------------
def read_txt_md(path: str) -> str:
with open(path, "r", encoding="utf-8", errors="ignore") as f:
return f.read()
def read_docx(path: str) -> str:
doc = DocxDocument(path)
return "\n".join(p.text for p in doc.paragraphs)
def read_pdf(path: str) -> str:
reader = PdfReader(path)
texts = []
for page in reader.pages:
try:
texts.append(page.extract_text() or "")
except Exception:
pass
return "\n".join(texts)
def read_csv_file(path: str, max_rows: int = 20000, max_chars: int = 400_000) -> str:
"""
Read a CSV as text that’s friendly for RAG.
- Coerces everything to string
- Skips bad lines
- Truncates very large files by rows and total chars (to avoid blowing up memory)
"""
try:
df = pd.read_csv(
path,
dtype=str,
on_bad_lines="skip", # pandas>=1.4
nrows=max_rows, # safeguard for very large CSVs
encoding="utf-8"
)
except UnicodeDecodeError:
# Fallback for odd encodings
df = pd.read_csv(
path,
dtype=str,
on_bad_lines="skip",
nrows=max_rows,
encoding="latin-1"
)
df = df.fillna("")
# Build a compact, readable text block:
# 1) schema
cols = list(df.columns)
schema_block = "COLUMNS: " + ", ".join(cols)
# 2) first N rows as pipe-separated lines (more RAG-friendly than raw CSV)
# Example line: "col1=..., col2=..., col3=..."
lines = []
for _, row in df.iterrows():
kv = [f"{c}={row[c]}" for c in cols]
lines.append(" | ".join(kv))
# Hard cap to avoid giant strings
if sum(len(x) for x in lines) > max_chars:
lines.append("…(truncated)")
break
data_block = "\n".join(lines)
return f"{schema_block}\n{data_block}"
def load_all_docs(folder: str) -> List[Tuple[str, str]]:
paths = []
# ADD "*.csv" here
for ext in ("*.md", "*.txt", "*.docx", "*.pdf", "*.csv"):
paths.extend(glob.glob(os.path.join(folder, ext)))
docs = []
for p in paths:
if p.endswith((".md", ".txt")):
text = read_txt_md(p)
elif p.endswith(".docx"):
text = read_docx(p)
elif p.endswith(".pdf"):
text = read_pdf(p)
elif p.endswith(".csv"): # <--- NEW
text = read_csv_file(p)
else:
continue
docs.append((p, text))
return docs
# -----------------------------
# Chunk + Embed + Index
# -----------------------------
def chunk_text(text: str, source: str, chunk_size: int = 300, overlap: int = 80) -> List[dict]:
words = text.split()
chunks = []
i = 0
while i < len(words):
chunk_words = words[i:i+chunk_size]
chunk = " ".join(chunk_words)
chunks.append({
"source": source,
"chunk": chunk,
"hash": hashlib.md5((source + str(i) + chunk).encode("utf-8")).hexdigest()
})
i += (chunk_size - overlap)
return chunks
def embed_texts(texts: List[str]) -> np.ndarray:
# Returns an array of shape (n, d)
# Uses OpenAI embeddings
resp = client.embeddings.create(model=EMBEDDING_MODEL, input=texts)
vecs = [item.embedding for item in resp.data]
return np.array(vecs).astype("float32")
@dataclass
class RAGIndex:
index: faiss.IndexFlatIP
df: pd.DataFrame # columns: [source, chunk, hash, vector]
dim: int
def build_or_load_index(force_rebuild: bool = False) -> RAGIndex:
docs = load_all_docs(DOCS_DIR)
if not docs:
raise RuntimeError(f"No docs found in {DOCS_DIR}/. Put your itinerary files there.")
# Simple staleness check: if any file is newer than index, rebuild
def newest_mtime():
paths = []
for ext in ("*.md", "*.txt", "*.docx", "*.pdf"):
paths.extend(glob.glob(os.path.join(DOCS_DIR, ext)))
return max(os.path.getmtime(p) for p in paths)
index_exists = os.path.exists(INDEX_PATH) and os.path.exists(META_CSV)
need_rebuild = force_rebuild
if index_exists:
idx_mtime = min(os.path.getmtime(INDEX_PATH), os.path.getmtime(META_CSV))
need_rebuild = need_rebuild or (newest_mtime() > idx_mtime)
if index_exists and not need_rebuild:
df = pd.read_csv(META_CSV)
vecs = np.load(INDEX_PATH)
dim = vecs.shape[1]
index = faiss.IndexFlatIP(dim)
faiss.normalize_L2(vecs)
index.add(vecs)
return RAGIndex(index=index, df=df, dim=dim)
# Rebuild
all_chunks = []
for path, text in docs:
if not text.strip():
continue
all_chunks.extend(chunk_text(text, source=path))
if not all_chunks:
raise RuntimeError("Docs were read but produced no chunks. Check formats.")
df = pd.DataFrame(all_chunks)
vecs = embed_texts(df["chunk"].tolist())
# Normalize for cosine similarity via inner product
faiss.normalize_L2(vecs)
np.save(INDEX_PATH, vecs)
df.to_csv(META_CSV, index=False)
dim = vecs.shape[1]
index = faiss.IndexFlatIP(dim)
index.add(vecs)
return RAGIndex(index=index, df=df, dim=dim)
# -----------------------------
# Retrieval + Answering
# -----------------------------
def retrieve(query: str, rag: RAGIndex, k: int = 7) -> List[dict]:
qvec = embed_texts([query])
faiss.normalize_L2(qvec)
D, I = rag.index.search(qvec, k)
results = []
for score, idx in zip(D[0], I[0]):
if idx == -1:
continue
row = rag.df.iloc[int(idx)]
results.append({
"score": float(score),
"source": row["source"],
"chunk": row["chunk"]
})
return results
SYSTEM_PROMPT = (
"You are a helpful, concise wedding assistant for Samuel's wedding. "
"Answer ONLY using the provided context from the wedding documents. "
"If the answer isn’t in the docs, say you don’t have that info and suggest who to contact (e.g., Overall IC). "
"Keep answers under 6 bullets or 150 words when possible. Use SGT times."
)
async def answer_with_rag(question: str, rag: RAGIndex) -> str:
ctx = retrieve(question, rag, k=6)
context_blocks = []
for r in ctx:
# Keep brief context slices
text = r["chunk"]
if len(text) > 800:
text = text[:800] + "…"
context_blocks.append(f"[Source: {os.path.basename(r['source'])}]\n{text}")
context_text = "\n\n".join(context_blocks)
messages = [
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": f"Context from docs:\n\n{context_text}\n\nQuestion: {question}"}
]
completion = client.chat.completions.create(
model=OPENAI_MODEL,
messages=messages,
temperature=0.2,
)
answer = completion.choices[0].message.content.strip()
if STRICT_DOC_MODE and (not context_blocks or "I don’t have that info" in answer):
# If no context matched strongly, enforce honesty
if len(context_blocks) == 0:
return (
"I couldn’t find this in the wedding docs. Please check the Family Playbook or ask the Overall IC. "
"You can also /refresh to make sure I have the latest files."
)
return answer
# -----------------------------
# Telegram Handlers
# -----------------------------
RAG = None # lazy loaded
async def ensure_rag(force: bool = False):
global RAG
if RAG is None or force:
RAG = build_or_load_index(force_rebuild=force)
async def start(update: Update, context: ContextTypes.DEFAULT_TYPE):
await ensure_rag(False)
msg = (
"👋 Hello! I’m the Wedding Q&A Bot. Ask me anything about roles, timings, addresses, and logistics.\n\n"
"Examples:\n"
"• What time is the solemnisation?\n"
"• What’s Mum’s role during tea ceremony?\n"
"• Where to park at the hotel?\n"
"• Who holds the ang bao box?\n\n"
"Admins can /refresh after updating the docs."
)
await update.message.reply_text(msg)
async def help_cmd(update: Update, context: ContextTypes.DEFAULT_TYPE):
await update.message.reply_text(
"Send a normal question, or use:\n"
"/role <name> — quick role lookup\n"
"/refresh — rebuild knowledge from latest docs (admin only, but not enforced)"
)
async def role_cmd(update: Update, context: ContextTypes.DEFAULT_TYPE):
await ensure_rag(False)
name = " ".join(context.args).strip()
if not name:
await update.message.reply_text("Usage: /role <name>")
return
q = f"What is the role and responsibilities of {name}? Include timings and contact if available."
ans = await answer_with_rag(q, RAG)
await update.message.reply_text(ans, parse_mode=ParseMode.MARKDOWN)
async def refresh_cmd(update: Update, context: ContextTypes.DEFAULT_TYPE):
await ensure_rag(True)
await update.message.reply_text("✅ Refreshed. I’m now using the latest documents in wedding_docs/.")
async def on_message(update: Update, context: ContextTypes.DEFAULT_TYPE):
await ensure_rag(False)
text = (update.message.text or "").strip()
if not text:
return
ans = await answer_with_rag(text, RAG)
# Telegram has 4096 char limit per message; be safe
if len(ans) > 3500:
ans = ans[:3500] + "…"
await update.message.reply_text(ans, parse_mode=ParseMode.MARKDOWN)
def main():
if not TELEGRAM_BOT_TOKEN:
raise RuntimeError("TELEGRAM_BOT_TOKEN missing")
if not OPENAI_API_KEY:
raise RuntimeError("OPENAI_API_KEY missing")
app = Application.builder().token(TELEGRAM_BOT_TOKEN).build()
app.add_handler(CommandHandler("start", start))
app.add_handler(CommandHandler("help", help_cmd))
app.add_handler(CommandHandler("role", role_cmd))
app.add_handler(CommandHandler("refresh", refresh_cmd))
app.add_handler(MessageHandler(filters.TEXT & ~filters.COMMAND, on_message))
print("Bot running… Press Ctrl+C to stop.")
app.run_polling(drop_pending_updates=True)
if __name__ == "__main__":
main() |