File size: 5,881 Bytes
82c35a1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env python3
"""
Dynamic RAG Database Updater
Processes PDFs in memory and updates the vector database.
No PDFs, OCR text, or intermediate files are stored.
"""

import numpy as np
from pathlib import Path
from typing import List, Dict
import pickle
from datetime import datetime

# PDF processing
from pdf2image import convert_from_path

# OCR
import pytesseract

# Embeddings
from sentence_transformers import SentenceTransformer

# FAISS
import faiss


class DynamicRAGUpdater:
    """
    Handles dynamic updates to RAG database:
    1. PDF upload (temporary path)
    2. OCR extraction (in memory)
    3. Chunking
    4. Embedding generation
    5. FAISS + metadata update ONLY
    """

    def __init__(
        self,
        vector_db_path: str,
        embedding_model: str = "microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext",
    ):
        self.vector_db_path = Path(vector_db_path)

        print("Using Tesseract OCR (in-memory only)")

        self.embedding_model = SentenceTransformer(
            embedding_model,
            device="cpu",
        )

        self.embedding_dim = self.embedding_model.get_sentence_embedding_dimension()
        self.load_database()

    # --------------------------------------------------
    # Load / Save database
    # --------------------------------------------------
    def load_database(self):
        index_file = self.vector_db_path / "faiss.index"
        metadata_file = self.vector_db_path / "metadata.pkl"

        self.faiss_index = faiss.read_index(str(index_file))

        with open(metadata_file, "rb") as f:
            data = pickle.load(f)
            self.chunks = data["chunks"]
            self.chunk_id_to_idx = data["chunk_id_to_idx"]

    def save_database(self):
        faiss.write_index(self.faiss_index, str(self.vector_db_path / "faiss.index"))

        with open(self.vector_db_path / "metadata.pkl", "wb") as f:
            pickle.dump(
                {
                    "chunks": self.chunks,
                    "chunk_id_to_idx": self.chunk_id_to_idx,
                    "embedding_dim": self.embedding_dim,
                    "model": "microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext",
                },
                f,
            )

    # --------------------------------------------------
    # OCR (in memory)
    # --------------------------------------------------
    def extract_text_from_pdf(self, pdf_path: str) -> str:
        try:
            images = convert_from_path(pdf_path, dpi=300)
        except Exception as e:
            raise RuntimeError(
                "PDF conversion failed. Ensure Poppler is installed."
            ) from e

        pages = []
        for page_num, image in enumerate(images, 1):
            text = pytesseract.image_to_string(
                image,
                lang="eng",
                config="--oem 3 --psm 6",
            )

            pages.append(
                f"\n{'=' * 40}\nPAGE {page_num}\n{'=' * 40}\n{text}"
            )

        return "\n".join(pages)

    # --------------------------------------------------
    # Chunking
    # --------------------------------------------------
    def chunk_text(self, text: str, chunk_size: int = 512) -> List[str]:
        sentences = text.split(". ")
        chunks, current, length = [], [], 0

        for s in sentences:
            s = s.strip()
            if not s:
                continue

            s += ". "
            if length + len(s) > chunk_size and current:
                chunks.append("".join(current))
                current = [s]
                length = len(s)
            else:
                current.append(s)
                length += len(s)

        if current:
            chunks.append("".join(current))

        return chunks

    # --------------------------------------------------
    # Embeddings
    # --------------------------------------------------
    def generate_embeddings(self, chunks: List[str]) -> np.ndarray:
        return self.embedding_model.encode(
            chunks,
            batch_size=32,
            convert_to_numpy=True,
            show_progress_bar=True,
        )

    # --------------------------------------------------
    # FAISS update
    # --------------------------------------------------
    def add_to_database(
        self,
        embeddings: np.ndarray,
        chunks: List[str],
        filename: str,
    ) -> int:
        start_idx = self.faiss_index.ntotal
        self.faiss_index.add(embeddings.astype("float32"))

        for i, text in enumerate(chunks):
            meta = {
                "chunk_id": start_idx + i,
                "text": text,
                "filename": filename,
                "upload_date": datetime.now().isoformat(),
                "source": "user_upload",
            }
            self.chunks.append(meta)
            self.chunk_id_to_idx[f"{filename}_{i}"] = start_idx + i

        return len(embeddings)

    # --------------------------------------------------
    # Full pipeline (NO FILE STORAGE)
    # --------------------------------------------------
    def process_and_add_pdf(self, pdf_path: str) -> Dict:
        start = datetime.now()
        filename = Path(pdf_path).stem

        # All steps are in memory
        text = self.extract_text_from_pdf(pdf_path)
        chunks = self.chunk_text(text)
        embeddings = self.generate_embeddings(chunks)

        vectors_added = self.add_to_database(embeddings, chunks, filename)
        self.save_database()

        return {
            "filename": filename,
            "num_chunks": len(chunks),
            "vectors_added": vectors_added,
            "total_vectors": self.faiss_index.ntotal,
            "processing_time_seconds": (datetime.now() - start).total_seconds(),
            "timestamp": datetime.now().isoformat(),
        }