Skip to content

LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows

[FEAT] Cell type merge - Table and record. (#1239)

harini-venkataramanbe56583684ef
diff --git a/backend/prompt_studio/prompt_studio_core_v2/constants.py b/backend/prompt_studio/prompt_studio_core_v2/constants.py
index ad3451175..1c13ca1e7 100644
--- a/backend/prompt_studio/prompt_studio_core_v2/constants.py
+++ b/backend/prompt_studio/prompt_studio_core_v2/constants.py
@@ -99,6 +99,7 @@ class ToolStudioPromptKeys:
ENABLE_HIGHLIGHT = "enable_highlight"
REQUIRED = "required"
EXECUTION_SOURCE = "execution_source"
+ LINE_ITEM = "line-item"
class FileViewTypes:
diff --git a/backend/prompt_studio/prompt_studio_core_v2/prompt_studio_helper.py b/backend/prompt_studio/prompt_studio_core_v2/prompt_studio_helper.py
index c90325100..056f3a785 100644
--- a/backend/prompt_studio/prompt_studio_core_v2/prompt_studio_helper.py
+++ b/backend/prompt_studio/prompt_studio_core_v2/prompt_studio_helper.py
@@ -981,7 +981,7 @@ class PromptStudioHelper:
tool_id: str,
output: dict[str, Any],
) -> dict[str, Any]:
- if prompt.enforce_type == TSPKeys.TABLE or prompt.enforce_type == TSPKeys.RECORD:
+ if prompt.enforce_type == TSPKeys.LINE_ITEM:
extract_doc_path: str = (
PromptStudioHelper._get_extract_or_summary_document_path(
org_id, user_id, tool_id, doc_name, TSPKeys.EXTRACT
diff --git a/backend/prompt_studio/prompt_studio_core_v2/static/select_choices.json b/backend/prompt_studio/prompt_studio_core_v2/static/select_choices.json
index 2e260a452..39265bece 100644
--- a/backend/prompt_studio/prompt_studio_core_v2/static/select_choices.json
+++ b/backend/prompt_studio/prompt_studio_core_v2/static/select_choices.json
@@ -8,14 +8,12 @@
"AZURE":"Azure OpenAI"
},
"output_type":{
- "string":"Text",
+ "string":"text",
"number":"number",
"email":"email",
"date":"date",
"boolean":"boolean",
"json":"json",
- "table":"table",
- "record":"record",
"line_item":"line-item"
},
"output_processing":{
diff --git a/backend/prompt_studio/prompt_studio_v2/migrations/0007_merge_table_record_cell_types.py b/backend/prompt_studio/prompt_studio_v2/migrations/0007_merge_table_record_cell_types.py
new file mode 100644
index 000000000..5f5ccd449
--- /dev/null
+++ b/backend/prompt_studio/prompt_studio_v2/migrations/0007_merge_table_record_cell_types.py
@@ -0,0 +1,46 @@
+# Generated by Django 4.2.1 on 2025-04-24 11:31
+
+from typing import Any
+
+from django.db import migrations, models
+
+
+def forward_enforce_type_updates(apps: Any, schema_editor: Any) -> None:
+ prompt = apps.get_model("prompt_studio_v2", "ToolStudioPrompt")
+
+ # line-item -> json
+ prompt.objects.filter(enforce_type="line-item").update(
+ enforce_type="json", is_line_item=True
+ )
+
+ # table or record -> line-item
+ prompt.objects.filter(enforce_type__in=["table", "record"]).update(
+ enforce_type="line-item"
+ )
+
+
+def revert_enforce_type_updates(apps, schema_editor: Any) -> None:
+ prompt = apps.get_model("prompt_studio_v2", "ToolStudioPrompt")
+
+ # Revert json -> line-item for records where is_line_item is True
+ prompt.objects.filter(enforce_type="json", is_line_item=True).update(
+ enforce_type="line-item"
+ )
+
+ # Revert line-item -> table or record for records
+ prompt.objects.filter(enforce_type="line-item").update(enforce_type="table")
+
+
+class Migration(migrations.Migration):
+ dependencies = [
+ ("prompt_studio_v2", "0006_alter_toolstudioprompt_enforce_type"),
+ ]
+
+ operations = [
+ migrations.AddField(
+ model_name="toolstudioprompt",
+ name="has_line_item_history",
+ field=models.BooleanField(default=False),
+ ),
+ migrations.RunPython(forward_enforce_type_updates, revert_enforce_type_updates),
+ ]
diff --git a/backend/prompt_studio/prompt_studio_v2/migrations/0008_alter_toolstudioprompt_enforce_type.py b/backend/prompt_studio/prompt_studio_v2/migrations/0008_alter_toolstudioprompt_enforce_type.py
new file mode 100644
index 000000000..4221de0f2
--- /dev/null
+++ b/backend/prompt_studio/prompt_studio_v2/migrations/0008_alter_toolstudioprompt_enforce_type.py
@@ -0,0 +1,33 @@
+# Generated by Django 4.2.1 on 2025-04-24 11:40
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+ dependencies = [
+ ("prompt_studio_v2", "0007_merge_table_record_cell_types"),
+ ]
+
+ operations = [
+ migrations.AlterField(
+ model_name="toolstudioprompt",
+ name="enforce_type",
+ field=models.TextField(
+ blank=True,
+ choices=[
+ ("Text", "Response sent as Text"),
+ ("number", "Response sent as number"),
+ ("email", "Response sent as email"),
+ ("date", "Response sent as date"),
+ ("boolean", "Response sent as boolean"),
+ ("json", "Response sent as json"),
+ (
+ "line-item",
+ "Response sent as line-item which is large a JSON output. If extraction stopped due to token limitation, we try to continue extraction from where it stopped",
+ ),
+ ],
+ db_comment="Field to store the type in which the response to be returned.",
+ default="Text",
+ ),
+ ),
+ ]
diff --git a/backend/prompt_studio/prompt_studio_v2/migrations/0009_alter_toolstudioprompt_enforce_type.py b/backend/prompt_studio/prompt_studio_v2/migrations/0009_alter_toolstudioprompt_enforce_type.py
new file mode 100644
index 000000000..37fdcd6c8
--- /dev/null
+++ b/backend/prompt_studio/prompt_studio_v2/migrations/0009_alter_toolstudioprompt_enforce_type.py
@@ -0,0 +1,33 @@
+# Generated by Django 4.2.1 on 2025-04-28 05:49
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+ dependencies = [
+ ("prompt_studio_v2", "0008_alter_toolstudioprompt_enforce_type"),
+ ]
+
+ operations = [
+ migrations.AlterField(
+ model_name="toolstudioprompt",
+ name="enforce_type",
+ field=models.TextField(
+ blank=True,
+ choices=[
+ ("text", "Response sent as Text"),
+ ("number", "Response sent as number"),
+ ("email", "Response sent as email"),
+ ("date", "Response sent as date"),
+ ("boolean", "Response sent as boolean"),
+ ("json", "Response sent as json"),
+ (
+ "line-item",
+ "Response sent as line-item which is large a JSON output. If extraction stopped due to token limitation, we try to continue extraction from where it stopped",
+ ),
+ ],
+ db_comment="Field to store the type in which the response to be returned.",
+ default="text",
+ ),
+ ),
+ ]
diff --git a/backend/prompt_studio/prompt_studio_v2/migrations/0010_update_enforce_type_text.py b/backend/prompt_studio/prompt_studio_v2/migrations/0010_update_enforce_type_text.py
new file mode 100644
index 000000000..3cb09d2ab
--- /dev/null
+++ b/backend/prompt_studio/prompt_studio_v2/migrations/0010_update_enforce_type_text.py
@@ -0,0 +1,27 @@
+# Generated by Django 4.2.1 on 2025-04-28 06:00
+
+from typing import Any
+
+from django.db import migrations
+
+
+def update_text_enforce_type(apps, schema_editor: Any) -> None:
+ prompt = apps.get_model("prompt_studio_v2", "ToolStudioPrompt")
+ prompt.objects.filter(enforce_type="Text").update(enforce_type="text")
+
+
+def revert_update_text_enforce_type(apps, schema_editor: Any) -> None:
+ prompt = apps.get_model("prompt_studio_v2", "ToolStudioPrompt")
+ prompt.objects.filter(enforce_type="text").update(enforce_type="Text")
+
+
+class Migration(migrations.Migration):
+ dependencies = [
+ ("prompt_studio_v2", "0009_alter_toolstudioprompt_enforce_type"),
+ ]
+
+ operations = [
+ migrations.RunPython(
+ update_text_enforce_type, reverse_code=revert_update_text_enforce_type
+ ),
+ ]
diff --git a/backend/prompt_studio/prompt_studio_v2/models.py b/backend/prompt_studio/prompt_studio_v2/models.py
index 6ec75f97e..402e79722 100644
--- a/backend/prompt_studio/prompt_studio_v2/models.py
+++ b/backend/prompt_studio/prompt_studio_v2/models.py
@@ -15,22 +15,12 @@ class ToolStudioPrompt(BaseModel):
"""
class EnforceType(models.TextChoices):
- TEXT = "Text", "Response sent as Text"
+ TEXT = "text", "Response sent as Text"
NUMBER = "number", "Response sent as number"
EMAIL = "email", "Response sent as email"
DATE = "date", "Response sent as date"
BOOLEAN = "boolean", "Response sent as boolean"
JSON = "json", "Response sent as json"
- TABLE = "table", "Response sent as table"
- RECORD = (
- "record",
- (
- "Response sent for records. "
- "Entries of records are list of "
- "logical and organized individual "
- "entities with distint values"
- ),
- )
LINE_ITEM = (
"line-item",
(
@@ -68,6 +58,8 @@ class ToolStudioPrompt(BaseModel):
choices=EnforceType.choices,
default=EnforceType.TEXT,
)
+ # New field to track if it was line-item
+ has_line_item_history = models.BooleanField(default=False)
prompt = models.TextField(
blank=True, db_comment="Field to store the prompt", unique=False
)
diff --git a/frontend/src/components/custom-tools/prompt-card/PromptCardItems.jsx b/frontend/src/components/custom-tools/prompt-card/PromptCardItems.jsx
index 3f3a25ba9..bb05367a4 100644
--- a/frontend/src/components/custom-tools/prompt-card/PromptCardItems.jsx
+++ b/frontend/src/components/custom-tools/prompt-card/PromptCardItems.jsx
@@ -19,7 +19,7 @@ import { useCustomToolStore } from "../../../store/custom-tool-store";
import { Header } from "./Header";
import { OutputForIndex } from "./OutputForIndex";
import { PromptOutput } from "./PromptOutput";
-import { TABLE_ENFORCE_TYPE, RECORD_ENFORCE_TYPE } from "./constants";
+import { LINE_ITEM_ENFORCE_TYPE } from "./constants";
let TableExtractionSettingsBtn;
try {
@@ -269,8 +269,7 @@ function PromptCardItems({
)}
</Space>
<Space>
- {(enforceType === TABLE_ENFORCE_TYPE ||
- enforceType === RECORD_ENFORCE_TYPE) &&
+ {enforceType === LINE_ITEM_ENFORCE_TYPE &&
TableExtractionSettingsBtn && (
<TableExtractionSettingsBtn
promptId={promptDetails?.prompt_id}
diff --git a/frontend/src/components/custom-tools/prompt-card/PromptOutput.jsx b/frontend/src/components/custom-tools/prompt-card/PromptOutput.jsx
index 29bda5fd5..71431e036 100644
--- a/frontend/src/components/custom-tools/prompt-card/PromptOutput.jsx
+++ b/frontend/src/components/custom-tools/prompt-card/PromptOutput.jsx
@@ -29,7 +29,7 @@ import {
import { TokenUsage } from "../token-usage/TokenUsage";
import { useWindowDimensions } from "../../../hooks/useWindowDimensions";
import { useCustomToolStore } from "../../../store/custom-tool-store";
-import { TABLE_ENFORCE_TYPE, RECORD_ENFORCE_TYPE } from "./constants";
+import { LINE_ITEM_ENFORCE_TYPE } from "./constants";
import { CopyPromptOutputBtn } from "./CopyPromptOutputBtn";
import { useAlertStore } from "../../../store/alert-store";
import { PromptOutputExpandBtn } from "./PromptOutputExpandBtn";
@@ -82,8 +82,7 @@ function PromptOutput({
} = useCustomToolStore();
const { setAlertDetails } = useAlertStore();
const { generatePromptOutputKey } = usePromptOutput();
- const isTableExtraction =
- enforceType === TABLE_ENFORCE_TYPE || enforceType === RECORD_ENFORCE_TYPE;
+ const isTableExtraction = enforceType === LINE_ITEM_ENFORCE_TYPE;
const noHighlightEnforceType = !["table", "record"].includes(enforceType);
const tooltipContent = (adapterConf) => (
<div>
diff --git a/frontend/src/components/custom-tools/prompt-card/PromptOutputsModal.jsx b/frontend/src/components/custom-tools/prompt-card/PromptOutputsModal.jsx
index 4fdce45b7..2f264283f 100644
--- a/frontend/src/components/custom-tools/prompt-card/PromptOutputsModal.jsx
+++ b/frontend/src/components/custom-tools/prompt-card/PromptOutputsModal.jsx
@@ -2,7 +2,7 @@ import { Col, Image, Modal, Row, Typography } from "antd";
import PropTypes from "prop-types";
import { DisplayPromptResult } from "./DisplayPromptResult";
-import { TABLE_ENFORCE_TYPE, RECORD_ENFORCE_TYPE } from "./constants";
+import { LINE_ITEM_ENFORCE_TYPE } from "./constants";
import SpaceWrapper from "../../widgets/space-wrapper/SpaceWrapper";
import { useCustomToolStore } from "../../../store/custom-tool-store";
import usePromptOutput from "../../../hooks/usePromptOutput";
@@ -81,9 +81,7 @@ function PromptOutputsModal({
)}
</div>
<div className="flex-1 overflow-y-auto pad-top-10">
- {(enforceType === TABLE_ENFORCE_TYPE ||
- enforceType === RECORD_ENFORCE_TYPE) &&
- TableOutput ? (
+ {enforceType === LINE_ITEM_ENFORCE_TYPE && TableOutput ? (
<TableOutput
output={promptOutputData?.output}
pagination={10}
diff --git a/frontend/src/components/custom-tools/prompt-card/constants.js b/frontend/src/components/custom-tools/prompt-card/constants.js
index 9e35aa5b0..cc6c7253d 100644
--- a/frontend/src/components/custom-tools/prompt-card/constants.js
+++ b/frontend/src/components/custom-tools/prompt-card/constants.js
@@ -1,5 +1,4 @@
-const TABLE_ENFORCE_TYPE = "table";
-const RECORD_ENFORCE_TYPE = "record";
+const LINE_ITEM_ENFORCE_TYPE = "line-item";
const handleUpdateStatus = (isUpdate, promptId, value, setUpdateStatus) => {
if (!isUpdate) {
@@ -11,4 +10,4 @@ const handleUpdateStatus = (isUpdate, promptId, value, setUpdateStatus) => {
});
};
-export { TABLE_ENFORCE_TYPE, handleUpdateStatus, RECORD_ENFORCE_TYPE };
+export { handleUpdateStatus, LINE_ITEM_ENFORCE_TYPE };
diff --git a/prompt-service/pyproject.toml b/prompt-service/pyproject.toml
index b3894b1ba..cc2757859 100644
--- a/prompt-service/pyproject.toml
+++ b/prompt-service/pyproject.toml
@@ -13,6 +13,7 @@ dependencies = [
"flask~=3.0",
"llama-index==0.12.8",
"python-dotenv==1.0.0",
+ "json-repair~=0.42.0",
# TODO: Temporarily removing the extra dependencies of aws and gcs from unstract-sdk
# to resolve lock file. Will have to be re-looked into
"unstract-sdk~=0.70.0",
diff --git a/prompt-service/src/unstract/prompt_service/constants.py b/prompt-service/src/unstract/prompt_service/constants.py
index 2ad922183..b49aec238 100644
--- a/prompt-service/src/unstract/prompt_service/constants.py
+++ b/prompt-service/src/unstract/prompt_service/constants.py
@@ -92,6 +92,7 @@ class PromptServiceConstants:
"provider to confirm writes to DB. "
"Please try again after some time"
)
+ COMBINED_PROMPT = "combined_prompt"
class RunLevel(Enum):
diff --git a/prompt-service/src/unstract/prompt_service/controllers/answer_prompt.py b/prompt-service/src/unstract/prompt_service/controllers/answer_prompt.py
index db950ac11..1fad52806 100644
--- a/prompt-service/src/unstract/prompt_service/controllers/answer_prompt.py
+++ b/prompt-service/src/unstract/prompt_service/controllers/answer_prompt.py
@@ -1,7 +1,5 @@
"""Published API Controller"""
-import json
-from json import JSONDecodeError
from typing import Any
from flask import Blueprint, request
@@ -173,13 +171,12 @@ def prompt_processor() -> Any:
)
return APIError(message=msg)
- if output[PSKeys.TYPE] == PSKeys.TABLE or output[PSKeys.TYPE] == PSKeys.RECORD:
+ if output[PSKeys.TYPE] == PSKeys.LINE_ITEM:
try:
- structured_output = AnswerPromptService.extract_table(
+ structured_output = AnswerPromptService.extract_line_item(
output=output,
structured_output=structured_output,
llm=llm,
- enforce_type=output[PSKeys.TYPE],
execution_source=execution_source,
)
metadata = UsageHelper.query_usage_metadata(
@@ -209,36 +206,6 @@ def prompt_processor() -> Any:
"Error while extracting table for the prompt",
)
raise api_error
- elif output[PSKeys.TYPE] == PSKeys.LINE_ITEM:
- try:
- structured_output = AnswerPromptService.extract_line_item(
- tool_settings=tool_settings,
- output=output,
- structured_output=structured_output,
- llm=llm,
- file_path=file_path,
- metadata=metadata,
- execution_source=execution_source,
- )
- continue
- except APIError as e:
- app.logger.error(
- "Failed to extract line-item for the prompt %s: %s",
- output[PSKeys.NAME],
- str(e),
- )
- publish_log(
- log_events_id,
- {
- "tool_id": tool_id,
- "prompt_key": prompt_name,
- "doc_name": doc_name,
- },
- LogLevel.ERROR,
- RunLevel.RUN,
- "Error while extracting line-item for the prompt",
- )
- raise e
try:
answer = "NA"
@@ -383,46 +350,19 @@ def prompt_processor() -> Any:
else:
structured_output[output[PSKeys.NAME]] = False
elif output[PSKeys.TYPE] == PSKeys.JSON:
- if answer.lower() == "[]" or answer.lower() == "na":
- structured_output[output[PSKeys.NAME]] = None
- else:
- try:
- structured_output[output[PSKeys.NAME]] = json.loads(answer)
- except JSONDecodeError:
- prompt = f"Convert the following text into valid JSON string: \
- \n{answer}\n\n The JSON string should be able to be parsed \
- into a Python dictionary. \
- Output just the JSON string. No explanation is required. \
- If you cannot extract the JSON string, output {{}}"
- try:
- answer = AnswerPromptService.run_completion(
- llm=llm,
- prompt=prompt,
- prompt_type=PSKeys.JSON,
- )
- structured_output[output[PSKeys.NAME]] = json.loads(answer)
- except JSONDecodeError as e:
- err_msg = (
- f"Error parsing response (to json): {e}\n"
- f"Candidate JSON: {answer}"
- )
- app.logger.info(err_msg, LogLevel.ERROR)
- # Format log message after unifying these types
- publish_log(
- log_events_id,
- {
- "tool_id": tool_id,
- "prompt_key": prompt_name,
- "doc_name": doc_name,
- },
- LogLevel.INFO,
- RunLevel.RUN,
- "Unable to parse JSON response from LLM, try using our"
- " cloud / enterprise feature of 'line-item', "
- "'record' or 'table' type",
- )
- structured_output[output[PSKeys.NAME]] = {}
-
+ AnswerPromptService.handle_json(
+ answer=answer,
+ structured_output=structured_output,
+ output=output,
+ log_events_id=log_events_id,
+ tool_id=tool_id,
+ doc_name=doc_name,
+ llm=llm,
+ enable_highlight=tool_settings.get(PSKeys.ENABLE_HIGHLIGHT, False),
+ execution_source=execution_source,
+ metadata=metadata,
+ file_path=file_path,
+ )
else:
structured_output[output[PSKeys.NAME]] = answer
diff --git a/prompt-service/src/unstract/prompt_service/services/answer_prompt.py b/prompt-service/src/unstract/prompt_service/services/answer_prompt.py
index 028099687..9770816cf 100644
--- a/prompt-service/src/unstract/prompt_service/services/answer_prompt.py
+++ b/prompt-service/src/unstract/prompt_service/services/answer_prompt.py
@@ -2,12 +2,16 @@ from logging import Logger
from typing import Any
from flask import current_app as app
+from json_repair import repair_json
from unstract.core.flask.exceptions import APIError
-from unstract.prompt_service.constants import ExecutionSource, FileStorageKeys
+from unstract.prompt_service.constants import ExecutionSource, FileStorageKeys, RunLevel
from unstract.prompt_service.constants import PromptServiceConstants as PSKeys
from unstract.prompt_service.exceptions import RateLimitError
from unstract.prompt_service.helpers.plugin import PluginManager
+from unstract.prompt_service.utils.file_utils import FileUtils
+from unstract.prompt_service.utils.log import publish_log
+from unstract.sdk.constants import LogLevel
from unstract.sdk.exceptions import RateLimitError as SdkRateLimitError
from unstract.sdk.exceptions import SdkError
from unstract.sdk.file_storage import FileStorage, FileStorageProvider
@@ -57,6 +61,9 @@ class AnswerPromptService:
enable_highlight = tool_settings.get(PSKeys.ENABLE_HIGHLIGHT, False)
if not enable_highlight or summarize_as_source:
platform_postamble = ""
+ plugin = PluginManager().get_plugin("json-extraction")
+ if plugin and hasattr(plugin["entrypoint_cls"], "update_settings"):
+ plugin["entrypoint_cls"].update_settings(tool_settings, output)
prompt = AnswerPromptService.construct_prompt(
preamble=tool_settings.get(PSKeys.PREAMBLE, ""),
prompt=output[prompt],
@@ -65,6 +72,7 @@ class AnswerPromptService:
context=context,
platform_postamble=platform_postamble,
)
+ output[PSKeys.COMBINED_PROMPT] = prompt
return AnswerPromptService.run_completion(
llm=llm,
prompt=prompt,
@@ -123,7 +131,11 @@ class AnswerPromptService:
PSKeys.HIGHLIGHT_DATA_PLUGIN
)
highlight_data = None
- if highlight_data_plugin and enable_highlight:
+ if (
+ highlight_data_plugin
+ and enable_highlight
+ and prompt_type.lower() != PSKeys.JSON
+ ):
fs_instance: FileStorage = FileStorage(FileStorageProvider.LOCAL)
if execution_source == ExecutionSource.IDE.value:
fs_instance = EnvHelper.get_storage(
@@ -142,7 +154,7 @@ class AnswerPromptService:
completion = llm.complete(
prompt=prompt,
process_text=highlight_data,
- extract_json=prompt_type.lower() != PSKeys.TEXT,
+ extract_json=prompt_type.lower() != PSKeys.JSON,
)
answer: str = completion[PSKeys.RESPONSE].text
highlight_data = completion.get(PSKeys.HIGHLIGHT_DATA, [])
@@ -169,18 +181,17 @@ class AnswerPromptService:
raise APIError(str(e)) from e
@staticmethod
- def extract_table(
+ def extract_line_item(
output: dict[str, Any],
structured_output: dict[str, Any],
llm: LLM,
- enforce_type: str,
execution_source: str,
) -> dict[str, Any]:
table_settings = output[PSKeys.TABLE_SETTINGS]
table_extractor: dict[str, Any] = PluginManager().get_plugin("table-extractor")
if not table_extractor:
raise APIError(
- "Unable to extract table details. "
+ "Unable to extract line-item details. "
"Please contact admin to resolve this issue."
)
fs_instance: FileStorage = FileStorage(FileStorageProvider.LOCAL)
@@ -198,7 +209,6 @@ class AnswerPromptService:
answer = table_extractor["entrypoint_cls"].extract_large_table(
llm=llm,
table_settings=table_settings,
- enforce_type=enforce_type,
fs_instance=fs_instance,
)
structured_output[output[PSKeys.NAME]] = answer
@@ -210,63 +220,91 @@ class AnswerPromptService:
raise APIError(message=msg)
@staticmethod
- def extract_line_item(
- tool_settings: dict[str, Any],
- output: dict[str, Any],
+ def handle_json(
+ answer: str,
structured_output: dict[str, Any],
+ output: dict[str, Any],
+ log_events_id: str,
+ tool_id: str,
+ doc_name: str,
llm: LLM,
- file_path: str,
- metadata: dict[str, str] | None,
- execution_source: str,
- ) -> dict[str, Any]:
- line_item_extraction_plugin: dict[str, Any] = PluginManager().get_plugin(
- "line-item-extraction"
- )
- if not line_item_extraction_plugin:
- raise APIError(PSKeys.PAID_FEATURE_MSG)
-
- extract_file_path = file_path
-
- # Read file content into context
- fs_instance: FileStorage = FileStorage(FileStorageProvider.LOCAL)
- if execution_source == ExecutionSource.IDE.value:
- fs_instance = EnvHelper.get_storage(
- storage_type=StorageType.PERMANENT,
- env_name=FileStorageKeys.PERMANENT_REMOTE_STORAGE,
- )
- if execution_source == ExecutionSource.TOOL.value:
- fs_instance = EnvHelper.get_storage(
- storage_type=StorageType.SHARED_TEMPORARY,
- env_name=FileStorageKeys.TEMPORARY_REMOTE_STORAGE,
- )
-
- if not fs_instance.exists(extract_file_path):
- raise FileNotFoundError(
- f"The file at path '{extract_file_path}' does not exist."
+ enable_highlight: bool = False,
+ execution_source: str = ExecutionSource.IDE.value,
+ metadata: dict[str, Any] | None = None,
+ file_path: str = "",
+ ) -> None:
+ """Handle JSON responses from the LLM."""
+ prompt_key = output[PSKeys.NAME]
+ if answer.lower() == "na":
+ structured_output[prompt_key] = None
+ else:
+ json_extraction_plugin = PluginManager().get_plugin("json-extraction")
+ if json_extraction_plugin:
+ answer = json_extraction_plugin["entrypoint_cls"](
+ llm=llm,
+ output=output,
+ prompt=output[PSKeys.COMBINED_PROMPT],
+ structured_output=structured_output,
+ answer=answer,
+ ).run()
+ if enable_highlight:
+ AnswerPromptService.handle_highlight(
+ execution_source=execution_source,
+ output=output,
+ answer=answer,
+ metadata=metadata,
+ file_path=file_path,
+ )
+ parsed_data = repair_json(
+ json_str=answer,
+ return_objects=True,
)
- context = fs_instance.read(path=extract_file_path, encoding="utf-8", mode="r")
-
- prompt = AnswerPromptService.construct_prompt(
- preamble=tool_settings.get(PSKeys.PREAMBLE, ""),
- prompt=output["promptx"],
- postamble=tool_settings.get(PSKeys.POSTAMBLE, ""),
- grammar_list=tool_settings.get(PSKeys.GRAMMAR, []),
- context=context,
- platform_postamble="",
- )
+ if isinstance(parsed_data, str):
+ err_msg = "Error parsing response (to json)\n" f"Candidate JSON: {answer}"
+ app.logger.info(err_msg, LogLevel.ERROR)
+ publish_log(
+ log_events_id,
+ {
+ "tool_id": tool_id,
+ "prompt_key": prompt_key,
+ "doc_name": doc_name,
+ },
+ LogLevel.INFO,
+ RunLevel.RUN,
+ "Unable to parse JSON response from LLM, try using our"
+ " cloud / enterprise feature 'record' or 'table' type",
+ )
+ structured_output[prompt_key] = {}
+ else:
+ structured_output[prompt_key] = parsed_data
- try:
- line_item_extraction = line_item_extraction_plugin["entrypoint_cls"](
- llm=llm,
- tool_settings=tool_settings,
- output=output,
- prompt=prompt,
- structured_output=structured_output,
- )
- answer = line_item_extraction.run()
- structured_output[output[PSKeys.NAME]] = answer
- metadata[PSKeys.CONTEXT][output[PSKeys.NAME]] = [context]
- return structured_output
- except line_item_extraction_plugin["exception_cls"] as e:
- msg = f"Couldn't extract table. {e}"
- raise APIError(message=msg)
+ @staticmethod
+ def handle_highlight(
+ execution_source: str,
+ output: dict[str, Any],
+ answer: str,
+ metadata: dict[str, Any] | None = None,
+ file_path: str = "",
+ ) -> None:
+ """Handle highlight data from the LLM."""
+ highlight_data_plugin = PluginManager().get_plugin("highlight-data")
+ if highlight_data_plugin:
+ highlight_response = highlight_data_plugin["entrypoint_cls"](
+ file_path=file_path,
+ fs_instance=FileUtils.get_fs_instance(execution_source=execution_source),
+ ).run(None, True, answer)
+ highlight_data = highlight_response.get(PSKeys.HIGHLIGHT_DATA, [])
+ confidence_data = highlight_response.get(PSKeys.CONFIDENCE_DATA)
+ line_numbers = highlight_response.get(PSKeys.LINE_NUMBERS, [])
+ whisper_hash = highlight_response.get(PSKeys.WHISPER_HASH, "")
+ prompt_key = output[PSKeys.NAME]
+ if metadata is not None and prompt_key:
+ metadata.setdefault(PSKeys.HIGHLIGHT_DATA, {})[prompt_key] = (
+ highlight_data
+ )
+ metadata.setdefault(PSKeys.LINE_NUMBERS, {})[prompt_key] = line_numbers
+ metadata[PSKeys.WHISPER_HASH] = whisper_hash
+ if confidence_data:
+ metadata.setdefault(PSKeys.CONFIDENCE_DATA, {})[prompt_key] = (
+ confidence_data
+ )
diff --git a/prompt-service/uv.lock b/prompt-service/uv.lock
index 3c4751da5..5b134d083 100644
--- a/prompt-service/uv.lock
+++ b/prompt-service/uv.lock
@@ -1169,6 +1169,15 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/91/29/df4b9b42f2be0b623cbd5e2140cafcaa2bef0759a00b7b70104dcfe2fb51/joblib-1.4.2-py3-none-any.whl", hash = "sha256:06d478d5674cbc267e7496a410ee875abd68e4340feff4490bcb7afb88060ae6", size = 301817 },
]
+[[package]]
+name = "json-repair"
+version = "0.42.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f4/1e/aea70e484e271e3d841773d5d185bedf488cf8f881088d47faae31f19116/json_repair-0.42.0.tar.gz", hash = "sha256:1a901f706c5b6b4325f0f79b53b0d998c5b327070e98b530da71cc5a3eda8616", size = 31349 }
+wheels = [
+ { url = "https://files.pythonhosted.org/packages/6a/c7/dd23f764de95771300a8a7ae17293c47ba0e48f826154229d18ecfe147cd/json_repair-0.42.0-py3-none-any.whl", hash = "sha256:7b6805162053dfe65722e961bc51b5eecec0582ec8a8e0fd218d33e8de757daf", size = 21612 },
+]
+
[[package]]
name = "jsonpath-python"
version = "1.0.6"
@@ -3002,6 +3011,7 @@ source = { editable = "." }
dependencies = [
{ name = "flask" },
{ name = "gcsfs" },
+ { name = "json-repair" },
{ name = "llama-index" },
{ name = "nltk" },
{ name = "peewee" },
@@ -3034,6 +3044,7 @@ test = [
requires-dist = [
{ name = "flask", specifier = "~=3.0" },
{ name = "gcsfs", specifier = "==2024.10.0" },
+ { name = "json-repair", specifier = "~=0.42.0" },
{ name = "llama-index", specifier = "==0.12.8" },
{ name = "nltk", specifier = "~=3.8" },
{ name = "peewee", specifier = "~=3.16" },