Field-Trusted GenAI Documentation

Tableau Workbook Knowledge Platform

Built an end-to-end documentation platform: parse TWB/TWBX into condensed metadata, two-pass Amazon Bedrock generation (draft + metadata-grounded refine), rule-based validator for calc/param/datasource coverage, versioned doc store (S3 pattern), and MCP-style retrieval tools. Local demo uses template mock; real Bedrock via shared portfolio_aws library + CDK-deployed Lambda and Secrets Manager.

★4/4 Validation Checks
Start a project like thisView source on GitHub

Challenge

Tableau workbooks accumulate calculated fields, parameters, and dashboards that only original authors understand—blocking self-service analytics, safe GenAI answers, and consistent metric definitions across teams.

Solution

Built an end-to-end documentation platform: parse TWB/TWBX into condensed metadata, two-pass Amazon Bedrock generation (draft + metadata-grounded refine), rule-based validator for calc/param/datasource coverage, versioned doc store (S3 pattern), and MCP-style retrieval tools. Local demo uses template mock; real Bedrock via shared portfolio_aws library + CDK-deployed Lambda and Secrets Manager.

Impact Metrics

4/4
Validation Checks
passed
2
Doc Passes
(draft + refine)
MCP
Retrieval
-style tools
CDK
Bedrock Path
+ Secrets Manager

Results

  • • 4/4 validation checks passed on demo workbook
  • • Two-pass doc generation with grounding against metadata.json
  • • MCP-style tools: list_workbooks, get_workbook_doc, search_docs
  • • CDK stack: API Gateway → Lambda → Bedrock Converse → S3
  • • No AWS credentials required for default local mock path

Business Impact

Turns tribal Tableau knowledge into field-trusted documentation that humans and agents can query—foundation for governed GenAI over BI assets.

Architecture

Tableau Workbook Knowledge PlatformTWB Parsemetadata.jsonBedrockValidatorS3 DocsMCPAWS CDK: API Gateway → Lambda → Bedrock → Secrets ManagerLocal mock: no AWS credentials • 4/4 validation passed

Sibling to Tableau → QuickSight Migration Assistant; shares metadata ingest layer.

Metadata → Trusted Docs

A proven approach combining statistical rigor, automation, and AWS best practices.

1

Parse & Condense

Extract data sources, calculated fields, parameters, sheets, and dashboards into minimal metadata.json.

2

Two-Pass Generation

Bedrock Converse draft, then refine pass constrained to metadata—reduces hallucinated field names.

3

Rule Validator

Non-LLM checks: calc coverage, param names, datasource references, required markdown sections.

4

Publish & Retrieve

Versioned S3-style layout; keyword search and MCP read tools for agents and analysts.

Scale & Scope

Demo
Validation
workbook — full pipeline under 2 minutes locally
4
Doc Passes
validation dimensions — calc, param, datasource, sections
CDK
AWS Path
PortfolioDemosStack — real Bedrock when deployed

Technology Stack

  • Python TWB XML parser
  • Amazon Bedrock Converse API
  • AWS CDK (Lambda, API Gateway, S3, Secrets Manager)
  • portfolio_aws shared library
  • MCP-style read server

Need a similar solution?

Let's replicate this success within your organization with a tailored engagement plan.

Start a ConversationView Our Process