jacob-mathison/doczyai-pipelines
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Activity

Revert

Browse Source
This commit is contained in:
Katon Minhas
2026-02-04 00:15:50 -05:00
parent 7e3d8ea2fe
commit 32c667ad49
1 changed files with 153 additions and 123 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/pipelines/shared/extraction/tin_npi_funcs.py
+153 -123
View File
@@ -2,16 +2,12 @@ import json
import logging import logging
import re import re
from typing import Optional
import src.constants.regex_patterns as regex_patterns import src.constants.regex_patterns as regex_patterns
import src.config as config import src.config as config
import src.prompts.prompt_templates as prompt_templates import src.prompts.prompt_templates as prompt_templates
from src.utils import llm_utils, string_utils from src.utils import llm_utils, string_utils
from src.pipelines.saas.prompts import prompt_calls from src.pipelines.saas.prompts import prompt_calls
from src.constants.delimiters import Delimiter
from src.prompts.fieldset import Field, FieldSet from src.prompts.fieldset import Field, FieldSet
from difflib import SequenceMatcher
def get_all_matches(text: str, pattern: str) -> list[str]: def get_all_matches(text: str, pattern: str) -> list[str]:
@@ -175,34 +171,38 @@ def clean_provider_info(provider_info: list[dict]) -> list[dict]:
for provider in provider_info: for provider in provider_info:
cleaned_provider = {} cleaned_provider = {}
# Clean TIN - remove hyphens and dots, ensure it's 9 digits # ---- Clean TIN (list, digits only, 9 chars) ----
if "TIN" in provider and provider["TIN"]: cleaned_tins = []
cleaned_provider["TIN"] = provider["TIN"].replace("-", "").replace(".", "") if isinstance(provider.get("TIN"), list):
# Validate that TIN only contains digits or "|" for tin in provider["TIN"]:
if not all(c.isdigit() or c == "|" for c in cleaned_provider["TIN"]): if not tin:
cleaned_provider["TIN"] = "UNKNOWN" continue
else: tin_str = str(tin).replace("-", "").replace(".", "")
cleaned_provider["TIN"] = "UNKNOWN" if tin_str.isdigit() and len(tin_str) == 9:
cleaned_tins.append(tin_str)
cleaned_provider["TIN"] = cleaned_tins if cleaned_tins else ["UNKNOWN"]
# Clean NPI - remove hyphens and dots, ensure it's 10 digits # ---- Clean NPI (list, digits only, 10 chars) ----
if "NPI" in provider and provider["NPI"]: cleaned_npis = []
cleaned_provider["NPI"] = provider["NPI"].replace("-", "").replace(".", "") if isinstance(provider.get("NPI"), list):
# Validate that NPI only contains digits or "|" for npi in provider["NPI"]:
if not all(c.isdigit() or c == "|" for c in cleaned_provider["NPI"]): if not npi:
cleaned_provider["NPI"] = "UNKNOWN" continue
else: npi_str = str(npi).replace("-", "").replace(".", "")
cleaned_provider["NPI"] = "UNKNOWN" if npi_str.isdigit() and len(npi_str) == 10:
cleaned_npis.append(npi_str)
cleaned_provider["NPI"] = cleaned_npis if cleaned_npis else ["UNKNOWN"]
# Clean Name - remove extra spaces and ensure it's not empty # ---- Clean NAME (list) ----
if "NAME" in provider and provider["NAME"]: cleaned_names = []
cleaned_provider["NAME"] = ( if isinstance(provider.get("NAME"), list):
" ".join(provider["NAME"].split()).strip().rstrip(".") for name in provider["NAME"]:
) if not name:
# If name is empty after cleaning, set to "UNKNOWN" continue
if not cleaned_provider["NAME"]: name_str = " ".join(str(name).split()).strip().rstrip(".")
cleaned_provider["NAME"] = "UNKNOWN" if name_str:
else: cleaned_names.append(name_str)
cleaned_provider["NAME"] = "UNKNOWN" cleaned_provider["NAME"] = cleaned_names if cleaned_names else ["UNKNOWN"]
cleaned_providers.append(cleaned_provider) cleaned_providers.append(cleaned_provider)
return cleaned_providers return cleaned_providers
@@ -220,7 +220,6 @@ def merge_provider_info_with_hybrid_smart_chunking(one_to_one_results, filename)
Returns: Returns:
dict: Updated one_to_one_results with merged values from provider_info. dict: Updated one_to_one_results with merged values from provider_info.
""" """
# Extract group and non-group providers # Extract group and non-group providers
group_tins = set() group_tins = set()
group_npis = set() group_npis = set()
@@ -229,48 +228,72 @@ def merge_provider_info_with_hybrid_smart_chunking(one_to_one_results, filename)
other_npis = [] other_npis = []
other_names = [] other_names = []
provider_name = one_to_one_results.get("PROVIDER_NAME") provider_name = one_to_one_results.get("PROVIDER_NAME") # Can be List or string
provider_info = one_to_one_results["PROV_INFO_JSON"] # Normalize provider_name to list: if string, convert to list to avoid character splitting
provider_list = ( if isinstance(provider_name, str):
provider_name = [provider_name]
elif not isinstance(provider_name, list):
provider_name = []
provider_info = one_to_one_results[
"PROV_INFO_JSON"
] # Returns string representation of list
prov_info_json_list = (
json.loads(provider_info) if isinstance(provider_info, str) else provider_info json.loads(provider_info) if isinstance(provider_info, str) else provider_info
) )
# Track if we found any name matches # Track if we found any name matches
found_match = False found_match = False
for provider in provider_list: for provider in prov_info_json_list:
# Ensure list format (FORMAT FIX ONLY)
tins = provider.get("TIN") or []
npis = provider.get("NPI") or []
names = provider.get("NAME") or []
# Normalize to lists: if string, convert to list to avoid character splitting
if isinstance(tins, str):
tins = [tins]
elif not isinstance(tins, list):
tins = []
if isinstance(npis, str):
npis = [npis]
elif not isinstance(npis, list):
npis = []
if isinstance(names, str):
names = [names]
elif not isinstance(names, list):
names = []
name_matches = prompt_calls.provider_name_match_check( name_matches = prompt_calls.provider_name_match_check(
provider_name, provider.get("NAME", "UNKNOWN"), filename names, provider_name, filename
) ) # Returns Boolean
if name_matches: if name_matches:
found_match = True found_match = True
provider["IS_GROUP"] = "Y" provider["IS_GROUP"] = "Y"
if provider.get("TIN"): for tin in tins:
group_tins.add(provider["TIN"]) group_tins.add(tin)
if provider.get("NPI"): for npi in npis:
group_npis.add(provider["NPI"]) group_npis.add(npi)
if provider.get("NAME"): for name in names:
group_names.add(provider["NAME"]) group_names.add(name)
else: else:
provider["IS_GROUP"] = "N" provider["IS_GROUP"] = "N"
if provider.get("TIN"): other_tins.extend(tins)
other_tins.append(provider["TIN"]) other_npis.extend(npis)
if provider.get("NPI"): other_names.extend(names)
other_npis.append(provider["NPI"])
if provider.get("NAME"):
other_names.append(provider["NAME"])
# If provider_name is not null and no match was found, add a new record # If provider_name is not null and no match was found, add a new record
if provider_name and not found_match: if provider_name and not found_match:
new_provider = { new_provider = {
"NAME": provider_name, "NAME": provider_name,
"TIN": "UNKNOWN", "TIN": [],
"NPI": "UNKNOWN", "NPI": [],
"IS_GROUP": "Y", "IS_GROUP": "Y",
} }
provider_list.append(new_provider) prov_info_json_list.append(new_provider)
group_names.add(provider_name) for name in provider_name:
group_names.add(name)
# De-Unknown GROUP fields # De-Unknown GROUP fields
group_tins = [v for v in group_tins if v != "UNKNOWN"] group_tins = [v for v in group_tins if v != "UNKNOWN"]
@@ -278,50 +301,42 @@ def merge_provider_info_with_hybrid_smart_chunking(one_to_one_results, filename)
group_names = [v for v in group_names if v != "UNKNOWN"] group_names = [v for v in group_names if v != "UNKNOWN"]
# Deduplicate while preserving order and remove any "UNKNOWN" entries and overlaps between group and other # Deduplicate while preserving order and remove any "UNKNOWN" entries and overlaps between group and other
group_tins = list(dict.fromkeys("|".join(group_tins).split("|"))) group_tins = list(dict.fromkeys(group_tins))
other_tins = list(dict.fromkeys("|".join(other_tins).split("|"))) other_tins = list(dict.fromkeys(other_tins))
other_tins = [ other_tins = [v for v in other_tins if v not in group_tins and v != "UNKNOWN"]
tin for tin in other_tins if tin not in group_tins and tin != "UNKNOWN"
] # remove any group TINs from other TINs
group_npis = list(dict.fromkeys("|".join(group_npis).split("|"))) group_npis = list(dict.fromkeys(group_npis))
other_npis = list(dict.fromkeys("|".join(other_npis).split("|"))) other_npis = list(dict.fromkeys(other_npis))
other_npis = [ other_npis = [v for v in other_npis if v not in group_npis and v != "UNKNOWN"]
npi for npi in other_npis if npi not in group_npis and npi != "UNKNOWN"
] # remove any group NPIs from other NPIs
group_names = list(dict.fromkeys("|".join(group_names).split("|"))) group_names = list(dict.fromkeys(group_names))
other_names = list(dict.fromkeys("|".join(other_names).split("|"))) other_names = list(dict.fromkeys(other_names))
other_names = [ other_names = [v for v in other_names if v not in group_names and v != "UNKNOWN"]
name for name in other_names if name not in group_names and name != "UNKNOWN"
] # remove any group Names from other Names
# Filter out records with NO_IDENTIFIERS_FOUND # Filter out records with NO_IDENTIFIERS_FOUND
provider_list = [ prov_info_json_list = [
p for p in provider_list if p.get("NAME") != "NO_IDENTIFIERS_FOUND" p
for p in prov_info_json_list
if "NO_IDENTIFIERS_FOUND" not in (p.get("NAME") or [])
] ]
# Deduplicate provider_list by NAME, keeping records with actual TIN/NPI values # Deduplicate provider_list by NAME, keeping records with actual TIN/NPI values
provider_list = deduplicate_providers_by_name(provider_list) prov_info_json_list = deduplicate_providers_by_name(prov_info_json_list)
# Reconstruct PROV_INFO_JSON and PROV_INFO_JSON_FORMATTED with IS_GROUP flags # Reconstruct PROV_INFO_JSON and PROV_INFO_JSON_FORMATTED with IS_GROUP flags
one_to_one_results["PROV_INFO_JSON"] = json.dumps(provider_list) one_to_one_results["PROV_INFO_JSON"] = json.dumps(prov_info_json_list)
one_to_one_results["PROV_INFO_JSON_FORMATTED"] = "\n".join( one_to_one_results["PROV_INFO_JSON_FORMATTED"] = "\n".join(
[json.dumps(provider) for provider in provider_list] [json.dumps(provider) for provider in prov_info_json_list]
) )
# Merge group provider information into one_to_one_results # Merge group provider information into one_to_one_results
one_to_one_results["PROV_GROUP_TIN"] = "|".join(group_tins) if group_tins else "" one_to_one_results["PROV_GROUP_TIN"] = group_tins if group_tins else []
one_to_one_results["PROV_GROUP_NPI"] = "|".join(group_npis) if group_npis else "" one_to_one_results["PROV_GROUP_NPI"] = group_npis if group_npis else []
one_to_one_results["PROV_GROUP_NAME_FULL"] = ( one_to_one_results["PROV_GROUP_NAME_FULL"] = group_names if group_names else []
"|".join(group_names) if group_names else ""
)
# Merge other provider information into one_to_one_results # Merge other provider information into one_to_one_results
one_to_one_results["PROV_OTHER_TIN"] = "|".join(other_tins) if other_tins else "" one_to_one_results["PROV_OTHER_TIN"] = other_tins if other_tins else []
one_to_one_results["PROV_OTHER_NPI"] = "|".join(other_npis) if other_npis else "" one_to_one_results["PROV_OTHER_NPI"] = other_npis if other_npis else []
one_to_one_results["PROV_OTHER_NAME_FULL"] = ( one_to_one_results["PROV_OTHER_NAME_FULL"] = other_names if other_names else []
"|".join(other_names) if other_names else ""
)
return one_to_one_results return one_to_one_results
@@ -414,30 +429,31 @@ def deduplicate_providers(
# Deduplicate if all three fields match # Deduplicate if all three fields match
for provider in provider_info: for provider in provider_info:
tins = provider.get("TIN", [])
npis = provider.get("NPI", [])
names = provider.get("NAME", [])
# Normalize missing values
tins = tins if isinstance(tins, list) else []
npis = npis if isinstance(npis, list) else []
names = names if isinstance(names, list) else []
# Skip providers where all critical identification fields are unknown # Skip providers where all critical identification fields are unknown
if ( if (
( (not tins or tins == ["UNKNOWN"])
string_utils.is_empty(provider.get("TIN", "")) and (not npis or npis == ["UNKNOWN"])
or provider.get("TIN", "") == "UNKNOWN" and (not names or names == ["UNKNOWN"])
)
and (
string_utils.is_empty(provider.get("NPI", ""))
or provider.get("NPI", "") == "UNKNOWN"
)
and (
string_utils.is_empty(provider.get("NAME", ""))
or provider.get("NAME", "") == "UNKNOWN"
)
): ):
continue continue
key = ( key = (
provider.get("TIN", ""), tuple(tins),
provider.get("NPI", ""), tuple(npis),
provider.get("NAME", ""), tuple(names),
) )
if key not in seen: if key not in seen:
seen.add(key) seen.add(key)
valid_providers.append(provider) valid_providers.append(provider)
return valid_providers return valid_providers
@@ -494,7 +510,7 @@ def get_provider_info(
logging.warning( logging.warning(
f"Warning: Unexpected format in provider info response: {type(providers)}" f"Warning: Unexpected format in provider info response: {type(providers)}"
) )
providers = [{"TIN": "UNKNOWN", "NPI": "UNKNOWN", "NAME": "PARSING_ERROR"}] providers = [{"TIN": [], "NPI": [], "NAME": ["PARSING_ERROR"]}]
# VALIDATION LAYER - Cross-check with regex findings # VALIDATION LAYER - Cross-check with regex findings
page_text = text_dict[page_num] page_text = text_dict[page_num]
@@ -523,19 +539,13 @@ def get_provider_info(
llm_found_tins = [] llm_found_tins = []
llm_found_npis = [] llm_found_npis = []
for provider in providers: for provider in providers:
if provider.get("TIN") and provider["TIN"] != "UNKNOWN": for tin in provider["TIN"] or []:
# Normalize for comparison (remove hyphens/dots) if tin and tin != "UNKNOWN":
raw_tins = [tin.strip() for tin in provider["TIN"].split("|")] llm_found_tins.append(tin.replace("-", "").replace(".", ""))
normalized_tins = [
tin.replace("-", "").replace(".", "") for tin in raw_tins for npi in provider["NPI"] or []:
] if npi and npi != "UNKNOWN":
llm_found_tins.extend(normalized_tins) llm_found_npis.append(npi.replace("-", "").replace(".", ""))
if provider.get("NPI") and provider["NPI"] != "UNKNOWN":
raw_npis = [npi.strip() for npi in provider["NPI"].split("|")]
normalized_npis = [
npi.replace("-", "").replace(".", "") for npi in raw_npis
]
llm_found_npis.extend(normalized_npis)
# Note: regex_tins and regex_npis are already normalized in get_all_matches_with_ocr # Note: regex_tins and regex_npis are already normalized in get_all_matches_with_ocr
missed_tins = [tin for tin in regex_tins if tin not in llm_found_tins] missed_tins = [tin for tin in regex_tins if tin not in llm_found_tins]
@@ -552,7 +562,7 @@ def get_provider_info(
logging.error(f"Error parsing JSON response from LLM: {str(e)}") logging.error(f"Error parsing JSON response from LLM: {str(e)}")
logging.error(f"Raw response: {llm_answer_raw}") logging.error(f"Raw response: {llm_answer_raw}")
# Return a default value # Return a default value
return [{"TIN": "UNKNOWN", "NPI": "UNKNOWN", "NAME": "PARSING_ERROR"}] return [{"TIN": [], "NPI": [], "NAME": ["PARSING_ERROR"]}]
def run_provider_info_fields( def run_provider_info_fields(
@@ -642,9 +652,9 @@ def run_provider_info_fields(
if not relevant_pages: if not relevant_pages:
deduplicated_provider_info = [ deduplicated_provider_info = [
{ {
"TIN": "UNKNOWN", "TIN": ["UNKNOWN"],
"NPI": "UNKNOWN", "NPI": ["UNKNOWN"],
"NAME": "NO_IDENTIFIERS_FOUND", "NAME": ["NO_IDENTIFIERS_FOUND"],
} }
] ]
else: else:
@@ -694,12 +704,32 @@ def deduplicate_providers_by_name(provider_list: list[dict]) -> list[dict]:
seen_names = set() seen_names = set()
deduplicated_list = [] deduplicated_list = []
# Sort so providers with actual TIN/NPI values come first # Sort so providers with actual TIN/NPI values come first
for provider in sorted( sorted_providers = sorted(
provider_list, provider_list,
key=lambda p: (p.get("TIN") == "UNKNOWN", p.get("NPI") == "UNKNOWN"), key=lambda p: (
): not p.get("TIN") or p.get("TIN") == ["UNKNOWN"],
name = provider.get("NAME") not p.get("NPI") or p.get("NPI") == ["UNKNOWN"],
if name not in seen_names: ),
)
for provider in sorted_providers:
names = provider.get("NAME", [])
# Skip invalid NAME values
if not isinstance(names, list) or not names or names == ["UNKNOWN"]:
continue
# Check if any name was already seen
is_duplicate = False
for name in names:
if name in seen_names:
is_duplicate = True
break
if is_duplicate:
continue
# Mark all names as seen
for name in names:
seen_names.add(name) seen_names.add(name)
deduplicated_list.append(provider) deduplicated_list.append(provider)
return deduplicated_list return deduplicated_list