Organizational knowledge is the information an organization maintains in policies, procedures, guides, directories, and other records. AI assistants need access to this information to ground answers in current facts rather than plausible guesses from training data. Conversation history and user memory do not replace it. section 2.1.3 section 10.4
Unlike conversation messages retrieved by ID, organizational documents are often unstructured. Retrieval must find passages relevant to a natural-language question even when the document uses different wording. A Data Service turns documents into searchable indexes and hides the ingestion and retrieval infrastructure from applications. section 5.0 section 2.1.3
A production knowledge pipeline separates several responsibilities:
- Ingest documents using format-specific parsers, then chunk the extracted text and generate embeddings. Chunk boundaries matter: splitting critical information can make a complete answer difficult to retrieve. section 5.7
- Configure indexes independently with their own embedding model, chunking strategy, and metadata schema. Replace old chunks when updated documents are reingested to keep indexed content current. section 5.7
- Return ranked results with document IDs and metadata. Hybrid search can combine vector and keyword results using Reciprocal Rank Fusion, avoiding score normalization. section 5.7
Healthcare intake
An intake assistant can search an index of provider procedures for a patient's question, pass retrieved text to the model, and return document IDs as sources. Current insurance guides and policies are more useful than manually maintained prompt content when information changes. section 2.1.3
Listing 2.3 Integrating organizational knowledge through Data Service
relevant_info = platform.data.search( #A
query=question, #A
index_name="patient_procedures" #A
)
relevant_text = "\n".join(r.text for r in relevant_info) #A
response = platform.models.chat(
model="gpt-4o",
system_prompt_name="patient_intake_assistant",
messages=[ #B
{"role": "system", #B
"content": f"Relevant information:\n{relevant_text}"}, #B
{"role": "user", "content": question}], #B
)
return {
"response": response.content,
"sources": [r.document_id for r in relevant_info] #C
}The listing shows the search-to-response flow, including how retrieved text and source document IDs reach the application. section 2.1.3
