iil-fieldprefill 0.2.0

AI field enrichment — shared prefill orchestration for Django platform apps

iil-fieldprefill

AI field enrichment — shared prefill orchestration for Django platform apps.

ADR-107: Use iil-fieldprefill as shared AI field enrichment layer.

Installation

pip install iil-fieldprefill                    # Core only
pip install 'iil-fieldprefill[django]'          # With Django HTMX mixin
pip install 'iil-fieldprefill[promptfw]'        # With promptfw templates
pip install 'iil-fieldprefill[all]'             # Everything

Quick Start

from fieldprefill import prefill_field, PrefillResult

result: PrefillResult = prefill_field(
    field_key="geltungsbereich",
    prompt="Beschreibe den Geltungsbereich...",
    action_code="ex_doc_prefill",
    max_tokens=2048,
)
print(result.content)        # Generated text
print(result.tokens_used)    # Input + output tokens
print(result.model)          # e.g. "gpt-4o"
print(result.latency_ms)     # Call duration

Multi-Field Prefill (v0.2.0+)

Enrich multiple fields in a single LLM call with structured JSON response:

from fieldprefill import prefill_fields

result = prefill_fields(
    field_keys=["description", "emotional_arc"],
    prompt="Enrich this chapter with a detailed description and emotional arc.",
    action_code="outline_enrich",
    context={"title": "Chapter 1", "act": "Act I", "beat_phase": "Inciting Incident"},
    scope="writing.outline_enrichment",
)

# Access individual fields from JSON response
node.description = result.get("description")
node.emotional_arc = result.get("emotional_arc")

# Or get full dict
data = result.as_dict()  # {"description": "...", "emotional_arc": "..."}

Cross-Hub Usage

risk-hub — Hazard field enrichment

# Register retrievers in AppConfig.ready()
from fieldprefill.retrievers import register_retriever

@register_retriever("sds")
def get_sds_texts(tenant_id, instance=None):
    from substances.models import Substance
    return [s.summary for s in Substance.objects.filter(tenant_id=tenant_id)]

# Prefill a field using existing document context
result = prefill_field(
    field_key="schutzmassnahmen",
    prompt="Beschreibe die erforderlichen Schutzmaßnahmen...",
    action_code="ex_doc_prefill",
    sources=["sds", "gefaehrdungsbeurteilung"],
    scope="explosionsschutz.ex_doc",
    tenant_id=request.tenant_id,
)

writing-hub — Outline node enrichment

@register_retriever("project_context")
def get_project_context(owner_id, instance=None):
    from apps.authoring.services.project_context_service import ProjectContextService
    ctx = ProjectContextService().get_context(str(instance.pk))
    return [ctx.to_prompt_block()]

result = prefill_fields(
    field_keys=["description", "emotional_arc"],
    prompt="Verfeinere dieses Kapitel mit detaillierter Beschreibung und emotionalem Bogen.",
    action_code="chapter_outline",
    sources=["project_context"],
    context={"title": node.title, "act": node.act, "beat_phase": node.beat_phase},
    scope="writing.outline_enrichment",
    tenant_id=user.pk,
    instance=node.outline_version.project,
)

Django HTMX Mixin

from fieldprefill.django import PrefillViewMixin

class MyEditView(LoginRequiredMixin, PrefillViewMixin, UpdateView):
    prefill_action_code = "ex_doc_prefill"
    prefill_scope = "explosionsschutz.ex_doc"

Architecture

App registers retrievers          fieldprefill orchestrates           aifw calls LLM
─────────────────────────    →    ─────────────────────────    →    ──────────────
@register_retriever("sds")        1. get_context_texts()             sync_completion()
@register_retriever("gbu")        2. build_messages()                completion()
register_system_prompt(...)       3. call_llm_sync/async()
                                  4. PrefillResult

License

MIT