Doubleword
    Back to home

    Workbooks

    Real examples with real costs. Each workbook links to runnable code.

    ExampleWhat It DoesCostKey Insight
    Async AgentsDeep research with recursive agent trees$0.34 for 47 agentsAsync SLA enables multi-round agentic workflows
    Synthetic Data GenerationGenerate training data with quality filtering$3.21 for 10K samples3-stage pipeline in 3 hours, not 3 days
    Data Processing PipelinesClean and enrich messy records$0.80 for 50K recordsLLM-powered normalization at pipeline scale
    EmbeddingsSemantic search over document corpus$0.03 for 1.6M tokens70% cheaper than OpenAI for same quality
    Async EvalsGrade LLM outputs with a frontier judge model$0.50 for 817 answers + judgementsRigorous evals as a background batch, not a hot-path cost
    LangSmith Evals at ScaleRun LangSmith LLM-as-judge evals via Doubleword's batch API$8.82 for 10K evals97% cheaper than GPT-5.5 real-time, zero 429 errors
    Bug Detection EnsembleClassify security vulnerabilities$0.40 for 4,642 samplesRun twice for calibration, still under $1
    Dataset LabellingBuild company datasets via search + LLM$1.05 for 188 companies100% recall vs Gartner Magic Quadrant
    Structured ExtractionExtract fields from scanned receipts$0.12 for 626 receiptsQwen3-30B beats GPT-5.2 at 1/12th cost
    Image SummarizationCaption images for social media$0.10 for 1,000 imagesVision batch makes captioning automatic
    AI Personal AssistantsAlways-on agents with background inference$0.80 for ~4M tokens/moBackground tasks at 4x lower cost than real-time
    Asynchronous Event ListenersProcess events asynchronously at scale$0.15 for 10K eventsDecouple event handling from request paths
    ETL & Pipeline SanitizationSanitize and transform data in ETL pipelines$0.50 for 100K recordsLLM-powered sanitization catches what regex misses
    Deep ResearchRecursive research agents over the web$0.34 for 47 agentsAsync SLA enables multi-round deep research
    Long Running AgentsAgents that run for hours, not seconds$1.20 for 8h runBackground inference unlocks long-horizon agents
    OpenClawOpen-source computer-use agent at batch cost$0.60 for 1K actionsComputer use without real-time premiums