File size: 6,489 Bytes
9b457ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Context building for RAG pipeline.

This module constructs optimized context from retrieved chunks for LLM consumption,
handling token limits and formatting.
"""

from typing import List, Optional
from dataclasses import dataclass
from src.config.settings import get_settings
from src.retrieval.retriever import RetrievedChunk
from src.utils.text_utils import count_tokens
from src.utils.logging import get_logger

logger = get_logger(__name__)


@dataclass
class RAGContext:
    """Context prepared for LLM with source information."""

    formatted_context: str
    sources: List[dict]
    total_tokens: int
    num_chunks: int


class ContextBuilder:
    """Build context from retrieved chunks for LLM prompts."""

    def __init__(self):
        """Initialize context builder with settings."""
        settings = get_settings()
        self.max_context_tokens = settings.max_context_tokens
        self.max_response_tokens = settings.max_response_tokens

    def build_context(
        self,
        chunks: List[RetrievedChunk],
        query: str,
        max_tokens: Optional[int] = None,
    ) -> RAGContext:
        """
        Build formatted context from retrieved chunks.

        Args:
            chunks: Retrieved chunks sorted by relevance
            query: Original user query
            max_tokens: Maximum tokens for context (default from settings)

        Returns:
            RAGContext: Formatted context with metadata
        """
        max_ctx_tokens = max_tokens or self.max_context_tokens

        # Reserve tokens for system prompt and response
        available_tokens = max_ctx_tokens - 2000  # Reserve for prompt overhead

        context_parts = []
        sources = []
        total_tokens = 0
        included_chunks = 0

        for chunk in chunks:
            # Check if we have room for this chunk
            chunk_tokens = chunk.token_count or count_tokens(chunk.text)

            if total_tokens + chunk_tokens > available_tokens:
                logger.debug(f"Token limit reached, stopping at {included_chunks} chunks")
                break

            # Format chunk with source citation
            chunk_text = self._format_chunk(chunk, included_chunks + 1)
            context_parts.append(chunk_text)

            # Track source with page info
            # Convert 0-indexed page numbers to 1-indexed for PDF.js
            page_1indexed = (chunk.page_numbers[0] + 1) if chunk.page_numbers else None
            source_type = chunk.source_type
            sources.append({
                "index": included_chunks + 1,
                "filename": chunk.filename,
                "chunk_id": chunk.chunk_id,
                "score": round(chunk.score, 3),
                "page_numbers": [p + 1 for p in chunk.page_numbers] if chunk.page_numbers else [],
                "page": page_1indexed,
                "source_type": source_type,
                "url": chunk.url if source_type != "local" else None,
            })

            total_tokens += chunk_tokens
            included_chunks += 1

        # Combine context
        formatted_context = "\n\n".join(context_parts)

        logger.info(f"Built context: {included_chunks} chunks, {total_tokens} tokens")

        return RAGContext(
            formatted_context=formatted_context,
            sources=sources,
            total_tokens=total_tokens,
            num_chunks=included_chunks,
        )

    def _format_chunk(self, chunk: RetrievedChunk, index: int) -> str:
        """
        Format a single chunk with source citation.

        Args:
            chunk: Retrieved chunk
            index: Citation index

        Returns:
            str: Formatted chunk text
        """
        source_type = chunk.source_type

        if source_type == "local":
            return f"[Source {index}: {chunk.filename}]\n{chunk.text}"
        elif source_type in ("duckduckgo", "tavily"):
            url = chunk.url or ""
            return f"[Source {index}: Web - {chunk.filename}]\nURL: {url}\n{chunk.text}"
        elif source_type in ("arxiv", "semantic_scholar", "pubmed"):
            url = chunk.url or ""
            return f"[Source {index}: Paper - {chunk.filename}]\nURL: {url}\n{chunk.text}"
        else:
            return f"[Source {index}: {chunk.filename}]\n{chunk.text}"

    def build_prompt(
        self,
        query: str,
        context: RAGContext,
        system_prompt: Optional[str] = None,
    ) -> dict:
        """
        Build the full prompt for the LLM.

        Args:
            query: User query
            context: RAG context with retrieved information
            system_prompt: Optional custom system prompt

        Returns:
            dict: Prompt structure with system and user messages
        """
        default_system = """You are a knowledgeable research assistant helping users understand documents in a PDF collection.

Your task is to answer questions based on the provided context from the documents. Follow these guidelines:

1. **Use the context**: Base your answers primarily on the information provided in the context sections.
2. **Cite sources**: When referencing information, cite the source using [Source N] format.
3. **Be accurate**: If the context doesn't contain enough information to fully answer the question, say so clearly.
4. **Be comprehensive**: Synthesize information from multiple sources when relevant.
5. **Be concise**: Provide clear, well-organized answers without unnecessary verbosity.

If the question cannot be answered from the provided context, explain what information is missing and suggest what might help."""

        system = system_prompt or default_system

        user_message = f"""Based on the following context from the document collection, please answer my question.

## Context from Documents

{context.formatted_context}

## Question

{query}

Please provide a comprehensive answer based on the context above, citing sources where appropriate."""

        return {
            "system": system,
            "user": user_message,
            "sources": context.sources,
        }

    def build_streaming_prompt(
        self,
        query: str,
        context: RAGContext,
    ) -> dict:
        """
        Build prompt optimized for streaming responses.

        Args:
            query: User query
            context: RAG context

        Returns:
            dict: Prompt structure for streaming
        """
        return self.build_prompt(query, context)