UWaterlooRFRChatBot / rag /prompt_builder.py
Anna Zhou
Add CS32 and source links to citations
0c2e8f6
Raw History Blame Contribute Delete
4.41 kB
import re
import json
import ast
_OVERVIEW_CONTEXT_CHARS = 1800
def _extract_label(text: str, label: str) -> str:
pattern = re.compile(
rf"\*\*{re.escape(label)}:\*\*\s*(.*?)(?=\n\n\*\*[^*\n]+:\*\*|\n---|\n#{1,6}\s|\Z)",
re.IGNORECASE | re.DOTALL,
)
match = pattern.search(text)
if not match:
return ""
return re.sub(r"\s+", " ", match.group(1)).strip()
def _compact_overview_text(text: str) -> str:
"""Keep overview prompts small enough that all cases fit in the answer."""
fields = []
for label in ("Case Name and Location", "Hazard(s)", "Type of Retreat", "Key Outcomes"):
value = _extract_label(text, label)
if value:
fields.append(f"{label}: {value}")
compact = "\n".join(fields)
if not compact:
compact = re.sub(r"\s+", " ", text).strip()
if len(compact) > _OVERVIEW_CONTEXT_CHARS:
compact = compact[:_OVERVIEW_CONTEXT_CHARS].rsplit(" ", 1)[0] + " ..."
return compact
def _parse_source_links(raw) -> list[dict]:
if not raw:
return []
if isinstance(raw, list):
return [x for x in raw if isinstance(x, dict)]
if isinstance(raw, str):
for parser in (json.loads, ast.literal_eval):
try:
parsed = parser(raw)
except Exception:
continue
if isinstance(parsed, list):
return [x for x in parsed if isinstance(x, dict)]
return []
def _format_source_links(metadata: dict, limit: int = 3) -> str:
links = _parse_source_links(metadata.get("source_links"))
pieces = []
for link in links[:limit]:
label = link.get("title") or link.get("doi") or "source"
url = link.get("url") or (f"https://doi.org/{link['doi']}" if link.get("doi") else "")
if url:
pieces.append(f"{label}: {url}")
return "; ".join(pieces)
def build_prompt(query: str, retrieved_chunks: list[dict], questionnaire_answers: dict | None = None) -> str:
is_overview = any(
kw in query.lower()
for kw in ("summar", "overview", "all case", "every case", "list", "compare", "across")
)
if is_overview:
context_blocks = "\n\n".join(
f"[Source: {c['metadata'].get('source', 'unknown')} | Case: {c['metadata'].get('case_id', '?')}]\n"
f"Source links: {_format_source_links(c['metadata'])}\n"
f"{_compact_overview_text(c['text'])}"
for c in retrieved_chunks
)
else:
context_blocks = "\n\n".join(
f"[Source: {c['metadata'].get('source', 'unknown')} | Case: {c['metadata'].get('case_id', '?')}]\n"
f"Source links: {_format_source_links(c['metadata'])}\n"
f"{c['text']}"
for c in retrieved_chunks
)
answers_section = ""
if questionnaire_answers:
answers_section = "\n\nUser context from questionnaire:\n" + "\n".join(
f"- {k}: {v}" for k, v in questionnaire_answers.items()
)
if is_overview:
case_ids = ", ".join(c["metadata"].get("case_id", "?") for c in retrieved_chunks)
instruction = (
"You are a research assistant. The user wants a summary of ALL case studies provided. "
f"The provided cases are: {case_ids}. "
"Write exactly one numbered bullet for EVERY provided case, in the same order. "
"Each bullet must start with the bold case ID, then give one concise sentence covering "
"location, hazard or retreat type, and the key outcome or lesson. "
"Keep each bullet under 45 words. Do not skip cases. Do not invent missing case IDs."
)
else:
instruction = (
"You are a research assistant helping municipal planners and community leaders understand managed retreat. "
"Answer in plain, formal language grounded strictly in the case study evidence below. "
"Do not use informal metaphors or colloquial expressions. "
"Cite specific case studies (e.g. CS5, CS20), name sources where possible, and include DOI/source links when they are supplied in the context. "
"Structure your answer with clear numbered sections if the question has multiple dimensions."
)
return f"""{instruction}{answers_section}
Case study excerpts:
{context_blocks}
Question: {query}
Answer:"""