Chunk PDF & Generate Embeddings for RAG | Azure AI Search Chunking API

RAG PIPELINE — STEP 1

Chunk and Perform Embedding

Upload a PDF to the Lowcademy Labs endpoint. It extracts the text, splits it into 800–1200-character semantic chunks, optionally calls Azure OpenAI Embeddings to generate a vector per chunk, and returns a payload you can POST directly to Azure AI Search without any modification. This is the first step in the RAG document ingestion pipeline.

POST · Lowcademy Hosted Azure OpenAI Embeddings · PDF Processing Postman Variable: openAI_Endpoint, openAI_Key, openAI_EmbeddingModelName

Endpoint

POST https://labs.lowcademy.com/apis/chunk-and-perform-embedding.php

Property	Value
Method	POST
Content-Type	multipart/form-data
Parameters	All parameters passed as request headers — not query params or form fields
File Upload	PDF binary in the `document` multipart form field
Response	Azure AI Search Documents–Index API payload: `{ "value": [...] }`

Postman Collection Variables Required

Set these variables in the Postman collection before calling this API These variables are only needed when performEmbedding: true. If you skip embedding, no Azure OpenAI variables are required.

openAI_Endpoint — e.g. https://my-resource.openai.azure.com/
openAI_Key — Your Azure OpenAI API key
openAI_EmbeddingModelName — e.g. text-embedding-3-small

Authentication

Lab API Key — For Training Use Only

Header: api-keyankitg.in

Header	Type	Required	Description
api-key	String	Required	Fixed lab API key. Returns `401` if missing or incorrect.

Request Headers

Header	Type	Required	Description
sourceName	String	Required	Logical name for the document (e.g. `AzureGuide`). Written to every chunk's `sourceName` field.
tenant_id	Int64	Required	Tenant identifier — must be numeric. Stored in every chunk.
document_id	Int64	Required	Document identifier — must be numeric. Used to build the chunk `id` key.
performEmbedding	Boolean	Optional	Pass `true` to generate Azure OpenAI embeddings per chunk. Defaults to `false`; embedding array will be `[]`.
embeddingEndpoint	String (URL)	Conditional	Your Azure OpenAI base URL. Required when `performEmbedding: true`. Maps to `{{openAI_Endpoint}}`.
embeddingKey	String	Conditional	Azure OpenAI API key. Required when `performEmbedding: true`. Maps to `{{openAI_Key}}`.
embeddingDeploymentName	String	Conditional	Embedding deployment name (e.g. `text-embedding-3-small`). Required when `performEmbedding: true`. Maps to `{{openAI_EmbeddingModelName}}`.

Request Body — File Upload

Field	Type	Required	Description
document	Binary (PDF)	Required	The PDF file. Must contain selectable text (not a scanned/image-only PDF). Send as `multipart/form-data`.

Sample Request

cURL — with embedding (embedding array populated)

curl -X POST \
  -H "api-key: ankitg.in" \
  -H "sourceName: MyDocument" \
  -H "tenant_id: 1" \
  -H "document_id: 42" \
  -H "performEmbedding: true" \
  -H "embeddingEndpoint: https://my-resource.openai.azure.com" \
  -H "embeddingKey: <your-azure-openai-key>" \
  -H "embeddingDeploymentName: text-embedding-3-small" \
  -F "document=@/path/to/document.pdf" \
  https://labs.lowcademy.com/apis/chunk-and-perform-embedding.php

cURL — without embedding (embedding will be [])

curl -X POST \
  -H "api-key: ankitg.in" \
  -H "sourceName: MyDocument" \
  -H "tenant_id: 1" \
  -H "document_id: 42" \
  -F "document=@/path/to/document.pdf" \
  https://labs.lowcademy.com/apis/chunk-and-perform-embedding.php

Response — 200 OK

Ready for Azure AI Search The response body is the exact JSON you POST to Azure AI Search /docs/index. No wrapping or transformation needed. Copy and send directly.

Field	Type	Description
value	Array	Array of chunk objects ready for Azure AI Search indexing.
value[].@search.action	String	Always `"upload"`.
value[].id	String	Unique key: `{tenant_id}-{document_id}-{zero-padded-index}`, e.g. `1-42-0001`.
value[].tenant_id	Int64	Echoed from the `tenant_id` header.
value[].document_id	Int64	Echoed from the `document_id` header.
value[].sourceName	String	Echoed from the `sourceName` header.
value[].section	String	Detected heading for this chunk (max 90 chars).
value[].content	String	Chunk text, 800–1200 characters.
value[].embedding	Array<float>	1536-dim float array when `performEmbedding: true`. Empty `[]` otherwise.

Sample Response

{
  "value": [
    {
      "@search.action" : "upload",
      "id"             : "1-42-0001",
      "tenant_id"      : 1,
      "document_id"    : 42,
      "sourceName"     : "MyDocument",
      "section"        : "Introduction",
      "content"        : "Azure AI Search is a fully managed cloud search service...",
      "embedding"      : [ 0.0023064255, -0.009327292, ... ]  // 1536 floats
    },
    // more chunks...
  ]
}

Error Responses

200

OK — Full Azure AI Search payload returned.

401

Unauthorized — api-key header missing or wrong.

400

Bad Request — Required header missing, tenant_id/document_id non-numeric, no file, or not a PDF. Also returned when performEmbedding: true but embedding headers are missing.

422

Unprocessable — No text extracted (scanned/image-only PDF).

502

Bad Gateway — Azure OpenAI embedding call failed. Includes the chunk index and upstream error.

405

Method Not Allowed — Request was not POST.

Next Step in RAG Pipeline

Step 2 — Index the response in Azure AI Search Take the full JSON response body from this API and POST it as-is to the Index Document API (Step 2):
POST https://<service>.search.windows.net/indexes/<index-name>/docs/index?api-version=2025-05-01-preview
with header api-key: <your-search-api-key> and Content-Type: application/json.