Skip to content

helpers

helpers

LLM helper utilities — sanitization, truncation, and factory.

truncate_legislative_text(text, verbatim_chars=LLM_VERBATIM_CHARS, max_chars=LLM_MAX_TEXT_CHARS)

Truncate Czech legislative text via structural extraction for LLM processing.

When TISK_SHORTENER is disabled (0), returns the full text unmodified.

Strategy (when enabled and text exceeds max_chars): 1. First verbatim_chars kept verbatim (covers explanatory report + most content) 2. From remainder: extract section headings + first 500 chars of each section 3. Hard cap at max_chars total

Source code in pspcz_analyzer/services/llm/helpers.py 
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
def truncate_legislative_text(
    text: str,
    verbatim_chars: int = LLM_VERBATIM_CHARS,
    max_chars: int = LLM_MAX_TEXT_CHARS,
) -> str:
    """Truncate Czech legislative text via structural extraction for LLM processing.

    When TISK_SHORTENER is disabled (0), returns the full text unmodified.

    Strategy (when enabled and text exceeds max_chars):
    1. First `verbatim_chars` kept verbatim (covers explanatory report + most content)
    2. From remainder: extract section headings + first 500 chars of each section
    3. Hard cap at `max_chars` total
    """
    if not TISK_SHORTENER:
        return text

    if len(text) <= max_chars:
        return text

    # Clamp verbatim to not exceed max (defensive for comparison calls)
    verbatim_chars = min(verbatim_chars, max_chars)

    result = text[:verbatim_chars]
    remainder = text[verbatim_chars:]

    # Extract structural highlights from the remainder
    highlights: list[str] = []
    for match in _HEADING_RE.finditer(remainder):
        start = match.start()
        snippet = remainder[start : start + 500 + len(match.group())]
        highlights.append(snippet.strip())

    if highlights:
        result += "\n\n[...]\n\n" + "\n\n".join(highlights)

    return result[:max_chars]

create_llm_client()

Factory: return the configured LLM backend client.

Reads LLM_PROVIDER from config: - "ollama" (default) -> LLMClient(provider="ollama", ...) - "openai" -> LLMClient(provider="openai", ...) (fails fast if OPENAI_API_KEY is empty)

Raises ValueError for unknown provider.

Source code in pspcz_analyzer/services/llm/helpers.py 
 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
def create_llm_client() -> LLMClient:
    """Factory: return the configured LLM backend client.

    Reads LLM_PROVIDER from config:
    - "ollama" (default) -> LLMClient(provider="ollama", ...)
    - "openai" -> LLMClient(provider="openai", ...) (fails fast if OPENAI_API_KEY is empty)

    Raises ValueError for unknown provider.
    """
    from pspcz_analyzer.services.llm.client import LLMClient

    match LLM_PROVIDER.lower().strip():
        case "ollama":
            return LLMClient(
                provider="ollama",
                base_url=OLLAMA_BASE_URL,
                model=OLLAMA_MODEL,
                api_key=OLLAMA_API_KEY,
                structured_output=LLM_STRUCTURED_OUTPUT,
            )
        case "openai":
            if not OPENAI_API_KEY:
                msg = (
                    "LLM_PROVIDER=openai but OPENAI_API_KEY is not set. "
                    "Set OPENAI_API_KEY in your .env or environment."
                )
                raise ValueError(msg)
            return LLMClient(
                provider="openai",
                base_url=OPENAI_BASE_URL,
                model=OPENAI_MODEL,
                api_key=OPENAI_API_KEY,
                structured_output=LLM_STRUCTURED_OUTPUT,
            )
        case _:
            msg = f"Unknown LLM_PROVIDER={LLM_PROVIDER!r}. Use 'ollama' or 'openai'."
            raise ValueError(msg)