tool-use-sequential

Tool Use Sequential: Search, Read, Extract, Summarize

A researcher needs an authoritative summary of a technical topic. A single web search returns a list of URLs, but URLs are not answers. The system needs to search, pick the best result, fetch the full page content, extract structured facts from the prose, and distill everything into a coherent summary with a confidence score.

This workflow chains four tools in strict sequence, where each tool's output becomes the next tool's input: search the web, read the top page, extract structured data, and produce a final summary.

Pipeline Architecture

query, maxResults
       |
       v
ts_search_web            (results list, topResult, totalFound=2450)
       |
       v
ts_read_page             (content with sections, statusCode=200, loadTimeMs=320)
       |
       v
ts_extract_data          (facts, keyFeatures, useCases, relevanceScore=0.92)
       |
       v
ts_summarize             (summary, confidence=0.94, wordCount)

Worker: SearchWeb (ts_search_web)

Returns three search results as List<Map<String, String>> with url, title, and snippet fields. Results point to Orkes and Netflix Conductor documentation. Exposes topResult separately for easy downstream wiring. Reports totalFound: 2450.

Worker: ReadPage (ts_read_page)

Reads the page at the given URL and returns a content map containing title, url, wordCount: 1850, and a sections list of four maps, each with heading and text fields covering "What is Conductor?", "Key Features", "Architecture", and "Use Cases". Reports statusCode: 200 and loadTimeMs: 320.

Worker: ExtractData (ts_extract_data)

Extracts structured data from page content. Returns a data map with facts (5 items about Conductor's origin, distributed execution, and fault tolerance), keyFeatures (5 items: task queuing, workflow versioning, error handling, visual designer, sync/async execution), useCases (5 items: media processing, microservice orchestration, data pipelines, CI/CD, AI/ML), origin: "Netflix", and type: "open-source". Reports relevanceScore: 0.92.

Worker: Summarize (ts_summarize)

Produces a coherent natural-language summary from the extracted data. Computes wordCount via split("\\s+").length on the summary text. Returns confidence: 0.94 and the sourceUrl for attribution.

Tests

4 tests cover web search, page reading, data extraction, and summarization.

Further Reading

• RUNNING.md -- how to build and run this example