LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows
Python ai-agentsdata-engineeringdocument-aigenerative-aiidpjson-extractionllmmcp-serverocrpdf-extractionprompt-engineeringstructured-output
[FIX] Fixed issue where summary context not being used properly (#1260)
Deepak K66dcbd2ca65e
diff --git a/backend/sample.env b/backend/sample.envindex c149e73e6..117a97e3a 100644--- a/backend/sample.env+++ b/backend/sample.env@@ -75,9 +75,9 @@ PROMPT_STUDIO_FILE_PATH=/app/prompt-studio-data# Structure Tool Image (Runs prompt studio exported tools)# https://hub.docker.com/r/unstract/tool-structure-STRUCTURE_TOOL_IMAGE_URL="docker:unstract/tool-structure:0.0.75"+STRUCTURE_TOOL_IMAGE_URL="docker:unstract/tool-structure:0.0.76"STRUCTURE_TOOL_IMAGE_NAME="unstract/tool-structure"-STRUCTURE_TOOL_IMAGE_TAG="0.0.75"+STRUCTURE_TOOL_IMAGE_TAG="0.0.76"# Feature FlagsEVALUATION_SERVER_IP=unstract-fliptdiff --git a/prompt-service/src/unstract/prompt_service/services/extraction.py b/prompt-service/src/unstract/prompt_service/services/extraction.pyindex 6d188425f..c6f0377b7 100644--- a/prompt-service/src/unstract/prompt_service/services/extraction.py+++ b/prompt-service/src/unstract/prompt_service/services/extraction.py@@ -86,7 +86,7 @@ class ExtractionService:metadata = {X2TextConstants.WHISPER_HASH: whisper_hash_value}for key, value in metadata.items():tool_exec_metadata[key] = value- metadata_path = str(Path(execution_run_data_folder / IKeys.METADATA_FILE))+ metadata_path = str(Path(execution_run_data_folder) / IKeys.METADATA_FILE)ToolUtils.dump_json(file_to_dump=metadata_path,json_to_dump=metadata,diff --git a/tools/structure/src/config/properties.json b/tools/structure/src/config/properties.jsonindex 9b618c952..dfa8e2142 100644--- a/tools/structure/src/config/properties.json+++ b/tools/structure/src/config/properties.json@@ -2,7 +2,7 @@"schemaVersion": "0.0.1","displayName": "Structure Tool","functionName": "structure_tool",- "toolVersion": "0.0.75",+ "toolVersion": "0.0.76","description": "This is a template tool which can answer set of input prompts designed in the Prompt Studio","input": {"description": "File that needs to be indexed and parsed for answers"diff --git a/tools/structure/src/main.py b/tools/structure/src/main.pyindex 0f0d79492..cc84e236f 100644--- a/tools/structure/src/main.py+++ b/tools/structure/src/main.py@@ -131,7 +131,7 @@ class StructureTool(BaseTool):if tool_settings[SettingsKeys.ENABLE_SINGLE_PASS_EXTRACTION]:if summarize_as_source:- summarize_file_hash = self._summarize_and_index(+ summarize_file_path, summarize_file_hash = self._summarize_and_index(tool_settings=tool_settings,tool_data_dir=tool_data_dir,responder=responder,@@ -139,6 +139,7 @@ class StructureTool(BaseTool):usage_kwargs=usage_kwargs,)payload[SettingsKeys.FILE_HASH] = summarize_file_hash+ payload[SettingsKeys.FILE_PATH] = summarize_file_pathself.stream_log("Fetching response for single pass extraction")# Since indexing is not involved for single passindex_metrics = {"time_taken(s)": 0}@@ -152,15 +153,18 @@ class StructureTool(BaseTool):reindex = Truefor output in outputs:if summarize_as_source:- summarize_file_hash = self._summarize_and_index(- tool_settings=tool_settings,- tool_data_dir=tool_data_dir,- responder=responder,- outputs=outputs,- usage_kwargs=usage_kwargs,+ summarize_file_path, summarize_file_hash = (+ self._summarize_and_index(+ tool_settings=tool_settings,+ tool_data_dir=tool_data_dir,+ responder=responder,+ outputs=outputs,+ usage_kwargs=usage_kwargs,+ ))payload[SettingsKeys.OUTPUTS] = outputspayload[SettingsKeys.FILE_HASH] = summarize_file_hash+ payload[SettingsKeys.FILE_PATH] = summarize_file_path# Since indexing is not involved for summaryindex_metrics[output[SettingsKeys.NAME]] = {"time_taken(s)": 0}break@@ -282,7 +286,7 @@ class StructureTool(BaseTool):responder: PromptTool,outputs: dict[str, Any],usage_kwargs: dict[Any, Any] = {},- ) -> str:+ ) -> tuple[str, str]:"""Summarizes the context of the file and indexes the summarizedcontent.@@ -345,11 +349,10 @@ class StructureTool(BaseTool):path=summarize_file_path, mode="w", data=summarized_context)- self.stream_log("Indexing summarized context")summarize_file_hash: str = self.workflow_filestorage.get_hash_from_file(path=summarize_file_path)- return summarize_file_hash+ return str(summarize_file_path), summarize_file_hashif __name__ == "__main__":