[DOC] Add Bedrock Cohere Rerank v3.5 connector blueprint

[gauravSsinha]

Description

Adds a connector blueprint for the Cohere Rerank v3.5 model on Amazon Bedrock (cohere.rerank-v3-5:0). This blueprint was missing from docs/remote_inference_blueprints/ despite the built-in pre/post processing functions being implemented in #3254.

Issues Resolved

Resolves #4831
Related: #3254, #3396

Changes Made

Added docs/remote_inference_blueprints/bedrock_connector_cohere_rerank_blueprint.md covering:

• Self-managed OpenSearch configuration with AWS credentials
• AWS OpenSearch Service configuration with IAM role ARN
• Built-in functions (connector.pre_process.cohere.rerank and connector.post_process.cohere.rerank) for OpenSearch 2.19+
• Custom Painless scripts for earlier OpenSearch versions (fallback)
• Model group creation, model registration, and deployment
• Inference testing via both Predict API and text similarity API
• Integration with OpenSearch reranking search pipeline
• query_text_path usage to avoid duplicate query specification

Context

The Cohere Rerank model on Bedrock is widely used for improving RAG pipeline relevance. The built-in pre/post processing functions were added in #3254 (closed Jan 2025), but no corresponding blueprint documentation was created. This PR fills that gap, following the same structure as existing blueprints (e.g., bedrock_connector_cohere_cohere.embed-english-v3_blueprint.md, bedrock_connector_titan_embedding_blueprint.md).

Testing

• Verified blueprint structure matches existing blueprints in the repository
• Connector configuration validated against the official OpenSearch reranking tutorial
• Built-in function names verified against source code in MLPreProcessFunction.java and MLPostProcessFunction.java

Checklist

• DCO sign-off on all commits
• Follows existing blueprint documentation format
• No code changes (documentation only)

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit 6050fa7)

Here are some key observations to aid the review process:

🧪 No relevant tests

🔒 Security concerns Sensitive information exposure: The blueprint includes placeholder text instructing users to insert AWS credentials (access_key, secret_key, session_token) directly into connector configurations (lines 36-38, 122-124). While a security note is present at line 63, it only appears after the first credential example and does not cover the second occurrence in section 2.3. Users following section 2.3 might miss the warning and commit credentials to version control or expose them in production configurations.

✅ No TODO sections

🔀 No multiple PR themes

⚡ No major issues detected

[github-actions]

PR Code Suggestions ✨

Latest suggestions up to 6050fa7
Explore these optional code suggestions:

Category	Suggestion	Impact
Possible issue	Add newline character escaping The Painless script does not handle newline characters in document text, which could break the JSON structure. Add escaping for newline (\n) and carriage return (\r) characters to prevent JSON parsing errors when documents contain line breaks. docs/remote_inference_blueprints/bedrock_connector_cohere_rerank_blueprint.md [142] -"pre_process_function": "\n def query_text = params.query_text;\n def text_docs = params.text_docs;\n def textDocsBuilder = new StringBuilder('[');\n for (int i=0; i<text_docs.length; i++) {\n textDocsBuilder.append('\"');\n textDocsBuilder.append(text_docs[i].replace('\\\\', '\\\\\\\\').replace('\"', '\\\\\"'));\n textDocsBuilder.append('\"');\n if (i<text_docs.length - 1) {\n textDocsBuilder.append(',');\n }\n }\n textDocsBuilder.append(']');\n def escapedQuery = query_text.replace('\\\\', '\\\\\\\\').replace('\"', '\\\\\"');\n def parameters = '{ \"query\": \"' + escapedQuery + '\", \"documents\": ' + textDocsBuilder.toString() + ' }';\n return '{\"parameters\": ' + parameters + '}';\n " +"pre_process_function": "\n def query_text = params.query_text;\n def text_docs = params.text_docs;\n def textDocsBuilder = new StringBuilder('[');\n for (int i=0; i<text_docs.length; i++) {\n textDocsBuilder.append('\"');\n textDocsBuilder.append(text_docs[i].replace('\\\\', '\\\\\\\\').replace('\"', '\\\\\"').replace('\\n', '\\\\n').replace('\\r', '\\\\r'));\n textDocsBuilder.append('\"');\n if (i<text_docs.length - 1) {\n textDocsBuilder.append(',');\n }\n }\n textDocsBuilder.append(']');\n def escapedQuery = query_text.replace('\\\\', '\\\\\\\\').replace('\"', '\\\\\"').replace('\\n', '\\\\n').replace('\\r', '\\\\r');\n def parameters = '{ \"query\": \"' + escapedQuery + '\", \"documents\": ' + textDocsBuilder.toString() + ' }';\n return '{\"parameters\": ' + parameters + '}';\n " Suggestion importance[1-10]: 8 __ Why: This is a critical bug fix. Without escaping newline and carriage return characters in the Painless script, documents containing line breaks will produce invalid JSON, causing parsing errors and connector failures. The suggestion correctly identifies the issue and provides the proper escaping for both query_text and text_docs.	Medium

---

Previous suggestions

Suggestions up to commit 1a5618e

Category	Suggestion	Impact
Security	Escape special characters in JSON The Painless script does not escape special characters in query_text or text_docs elements, which can break JSON formatting if documents contain quotes, newlines, or backslashes. This could cause request failures or injection vulnerabilities. Add proper JSON escaping for string values. docs/remote_inference_blueprints/bedrock_connector_cohere_rerank_blueprint.md [143] -"pre_process_function": "\n def query_text = params.query_text;\n def text_docs = params.text_docs;\n def textDocsBuilder = new StringBuilder('[');\n for (int i=0; i<text_docs.length; i++) {\n textDocsBuilder.append('\"');\n textDocsBuilder.append(text_docs[i]);\n textDocsBuilder.append('\"');\n if (i<text_docs.length - 1) {\n textDocsBuilder.append(',');\n }\n }\n textDocsBuilder.append(']');\n def parameters = '{ \"query\": \"' + query_text + '\", \"documents\": ' + textDocsBuilder.toString() + ' }';\n return '{\"parameters\": ' + parameters + '}';\n ", +"pre_process_function": "\n def query_text = params.query_text.replace('\\\\', '\\\\\\\\').replace('\"', '\\\\\"').replace('\\n', '\\\\n').replace('\\r', '\\\\r');\n def text_docs = params.text_docs;\n def textDocsBuilder = new StringBuilder('[');\n for (int i=0; i<text_docs.length; i++) {\n def escaped = text_docs[i].replace('\\\\', '\\\\\\\\').replace('\"', '\\\\\"').replace('\\n', '\\\\n').replace('\\r', '\\\\r');\n textDocsBuilder.append('\"');\n textDocsBuilder.append(escaped);\n textDocsBuilder.append('\"');\n if (i<text_docs.length - 1) {\n textDocsBuilder.append(',');\n }\n }\n textDocsBuilder.append(']');\n def parameters = '{ \"query\": \"' + query_text + '\", \"documents\": ' + textDocsBuilder.toString() + ' }';\n return '{\"parameters\": ' + parameters + '}';\n ", Suggestion importance[1-10]: 8 __ Why: This is a critical security and correctness issue. The Painless script in pre_process_function does not escape special characters like quotes, backslashes, or newlines in query_text or text_docs, which can break JSON formatting or introduce injection vulnerabilities when documents contain these characters.	Medium
General	Hardcode api_version in request body The request_body uses string interpolation for api_version without quotes, which will result in invalid JSON if the value is not a number. Since api_version is set to 2 in parameters, this works, but the template is fragile. Consider using proper JSON formatting or document this requirement. docs/remote_inference_blueprints/bedrock_connector_cohere_rerank_blueprint.md [56] -"request_body": "{ \"documents\": ${parameters.documents}, \"query\": \"${parameters.query}\", \"api_version\": ${parameters.api_version} }", +"request_body": "{ \"documents\": ${parameters.documents}, \"query\": \"${parameters.query}\", \"api_version\": 2 }", Suggestion importance[1-10]: 3 __ Why: While the suggestion correctly identifies that api_version is interpolated without quotes, the current approach is intentional and works correctly since api_version is set to 2 (a number). Hardcoding reduces flexibility without significant benefit, as the parameter is already defined in the configuration.	Low

[github-actions]

Persistent review updated to latest commit 6050fa7