harbor / src /utils /resources.py
Lyonel Tanganco
cleanup
d6c18ca
import csv
# Profile field → (csv_column, weight)
FIELD_MAP = [
("clinical", "primary_focus", "primary_focus", 3),
("clinical", "substances", "substances", 3),
("demographics", "population", "age_groups", 2),
("demographics", "identity_factors", "identity_factors", 2),
("logistics", "insurance", "insurance", 2),
("preferences", "setting", "settings", 2),
("preferences", "therapy_approach", "therapy_approaches", 1),
("demographics", "language", "languages", 1),
]
def load_resources(csv_path):
"""Load one or more resource CSVs into a list of dicts.
Accepts a single path (str) or a list of paths. Called once at init.
"""
if isinstance(csv_path, str):
csv_path = [csv_path]
rows = []
for path in csv_path:
with open(path, "r", encoding="utf-8") as f:
reader = csv.DictReader(f)
rows.extend(reader)
return rows
def _get_profile_value(profile, category, field):
"""Safely get a profile value, returning None for missing/empty."""
val = profile.get(category, {}).get(field)
if val is None:
return None
if isinstance(val, list) and len(val) == 0:
return None
return val
def _pipe_values(cell):
"""Split a pipe-delimited CSV cell into a set of lowercase values."""
if not cell or not cell.strip():
return set()
return {v.strip().lower() for v in cell.split("|")}
def filter_resources(resources, user_profile):
"""
Filter the full resource list down to a relevant subset based on
user profile values. Applies geographic, primary_focus, and substances
filters. Progressively relaxes filters if fewer than 3 results remain.
"""
zipcode = _get_profile_value(user_profile, "logistics", "zipcode")
region = _get_profile_value(user_profile, "logistics", "region")
primary_focus = _get_profile_value(user_profile, "clinical", "primary_focus")
substances = _get_profile_value(user_profile, "clinical", "substances")
# No profile info → no filtering possible, return empty (no recommendations)
if not zipcode and not region and not primary_focus and not substances:
return []
# Build filter functions in order of relaxation priority
filters = []
# Geographic filter (relaxed first if too few results)
if zipcode:
zip_prefix = zipcode[:3]
filters.append(("geo", lambda r, zp=zip_prefix: (
r.get("zip", "")[:3] == zp
)))
elif region:
region_lower = region.lower()
filters.append(("geo", lambda r, rl=region_lower: (
rl in r.get("city", "").lower() or rl in r.get("state", "").lower()
)))
# Primary focus filter
if primary_focus:
focus_lower = primary_focus.lower()
filters.append(("focus", lambda r, fl=focus_lower: (
not r.get("primary_focus", "").strip() or
fl in _pipe_values(r.get("primary_focus", ""))
)))
# Substances filter
if substances:
if isinstance(substances, str):
substances = [substances]
subs_lower = {s.lower() for s in substances}
filters.append(("substances", lambda r, sl=subs_lower: (
not r.get("substances", "").strip() or
bool(sl & _pipe_values(r.get("substances", "")))
)))
# Apply all filters, progressively relax if < 3 results
result = _apply_filters(resources, filters)
if len(result) >= 3:
return result
best = result # keep the best partial matches found so far
# Relax geographic filter first
relaxed = [f for f in filters if f[0] != "geo"]
if relaxed:
result = _apply_filters(resources, relaxed)
if len(result) >= 3:
return result
if len(result) > len(best):
best = result
# Relax substances filter next
relaxed = [f for f in relaxed if f[0] != "substances"]
if relaxed:
result = _apply_filters(resources, relaxed)
if len(result) > len(best):
best = result
return best
def _apply_filters(resources, filters):
"""Apply a list of filter functions, keeping rows that pass all."""
if not filters:
return []
result = []
for row in resources:
if all(fn(row) for _, fn in filters):
result.append(row)
return result
def score_resources(filtered, user_profile, top_n=3):
"""
Score filtered resources by relevance to the user profile.
Returns the top_n highest-scoring resources as a list of dicts.
"""
zipcode = _get_profile_value(user_profile, "logistics", "zipcode")
region = _get_profile_value(user_profile, "logistics", "region")
scored = []
for row in filtered:
score = 0
# Score each mapped field
for category, field, csv_col, weight in FIELD_MAP:
profile_val = _get_profile_value(user_profile, category, field)
if profile_val is None:
continue
cell_values = _pipe_values(row.get(csv_col, ""))
if not cell_values:
continue # empty cell = neutral
if isinstance(profile_val, list):
matches = sum(1 for v in profile_val if v.lower() in cell_values)
if matches > 0:
score += weight * (matches / len(profile_val))
else:
if profile_val.lower() in cell_values:
score += weight
# Geographic bonus
row_zip = row.get("zip", "").strip()
if zipcode and row_zip:
if row_zip == zipcode:
score += 5
elif row_zip[:3] == zipcode[:3]:
score += 2
elif region and not zipcode:
region_lower = region.lower()
if region_lower in row.get("city", "").lower():
score += 3
if score > 0:
scored.append((score, row))
# Sort by score descending
scored.sort(key=lambda x: x[0], reverse=True)
return [row for _, row in scored[:top_n]]
def format_resources_for_context(results):
"""
Format a list of resource dicts as a context block for injection into the
system prompt. The LLM uses this verified data to present recommendations
naturally in its own voice. Returns empty string if no results.
"""
if not results:
return ""
lines = [
"[VERIFIED FACILITY DATA — Present these facilities to the user following the "
"output format in your instructions. Use only the data listed here — do not invent, "
"alter, or supplement with facilities not in this list.]",
"",
]
for i, row in enumerate(results, 1):
name = row.get("name", "Unknown Facility")
lines.append(f"Facility {i}: {name}")
parts = [row.get("address", ""), row.get("city", ""),
row.get("state", ""), row.get("zip", "")]
address = ", ".join(p.strip() for p in parts if p.strip())
if address:
lines.append(f" Address: {address}")
phone = row.get("phone", "").strip()
if phone:
lines.append(f" Phone: {phone}")
website = row.get("website", "").strip()
if website:
lines.append(f" Website: {website}")
focus = row.get("primary_focus", "").strip()
if focus:
lines.append(" Focus: " + ", ".join(
v.strip().replace("_", " ").title() for v in focus.split("|")
))
subs = row.get("substances", "").strip()
if subs:
lines.append(" Substances: " + ", ".join(
v.strip().replace("_", " ").title() for v in subs.split("|")
))
settings = row.get("settings", "").strip()
if settings:
lines.append(" Settings: " + ", ".join(
v.strip().replace("_", " ").title() for v in settings.split("|")
))
insurance = row.get("insurance", "").strip()
if insurance:
lines.append(" Insurance: " + ", ".join(
v.strip().replace("_", " ").title() for v in insurance.split("|")
))
lines.append("")
return "\n".join(lines).rstrip()