Home › Simulators

Run it · step 2 of the loop

Simulators

Reading about a system is one thing; watching it work, step by step, and then breaking it on purpose is how it sticks. Every simulator here says whether its numbers were recorded for real or are illustrative, never passed off as measured.

AI agents

Watch an agent work, then watch it fail and see what catches it.

Claude API Simulator

one request, one response illustrative

Change the request (model, system prompt, max_tokens, effort, thinking, temperature, stop sequences, a tool, streaming) and watch the Python, TypeScript and cURL code, the JSON response or streamed events, the tokens and the cost update.

Agent Harness Simulator

the claims assistant, one turn at a time

Type a message and watch every step: the systems it talks to, the checks that fire, the code, the variables, the JSON to and from the model, and the trace.

AgentTwin: a tiny library agent

5 scenarios, a well-built and a faulty agent illustrative

Run each scenario against two agents and see which of four judges catches what: something owed and never done, a false reply, a leak, a planted instruction.

A day with Babuji

a companion for older adults illustrative

Twelve moments from one day with Babu, a Hindi-speaking companion: what it says, and the rules behind each reply.

Data platforms: the DataTwin Studio

Real runs on Apache Spark 3.5 and Delta Lake 3.2, recorded on a laptop, so you can see inside the engine.

One batch, end to end

scene 1 of the DataTwin Studio recorded for real

Follow one day of orders from the source file to the dashboard: config, cluster, every layer, the files on disk, the Delta log. Then load the same file again and watch.

The Problem Lab

19 problems every data team hits recorded for real

Duplicates, re-runs, SCD2 pitfalls, late data, failure mid-job, schema drift, deletes, skew, personal data. Each one: what you see, the naive fix, the right fix.

Performance stories

the slow job, the clue, the fix recorded for real

Detective cases from hours to minutes, each with the Spark UI and event-log clues and the before-and-after numbers.

Delta Lake internals

files, log, versions recorded for real

What a commit writes, how a version is read, time travel, what MERGE really rewrites, deletion vectors.

Spark internals

jobs, stages, tasks, plans recorded for real

From one query to the stages and tasks that ran it, with the query plan, shuffles and joins.

Data quality as metadata

one port, two engines recorded for real

Rules written as YAML, run by Spark SQL and by Deequ with the same verdicts; quarantine, profiling and an anomaly check.

Streaming simulator

IoT and database data, fast lane and medallion illustrative

Pick a binding (AWS + Databricks, AWS only, Databricks only) and a route (direct, via the medallion, both), watch data reach a live screen and Power BI, then break a gateway or a nightly batch. A second page runs fast, near-real-time and batch lanes over a simulated day.

DataTwin: a tiny platform

5 days, a well-built and a faulty platform illustrative

A day arrives twice, a column is renamed, a refund is negative, a real sale spike, a source goes dark. Is the dashboard still right?

Exam practice

Drills that put you in the trap before the exam does.

CCA-F Trap Gym

Claude Certified Architect

Interactive drills for the traps the exam sets: too many tools, a filling context window, a model reviewing its own work.