Latentlog
Notes from the machine room.
Technical writing on data transformation, rule inference, automation patterns, and the things we are learning while building Latentmachine.
Use Latentmachine as an MCP Server for Claude and Cursor
Connect Latentmachine to Claude Desktop, Claude Code, or Cursor as an MCP server. The AI does the data transformation, then calls a deterministic engine to verify whether every row is consistent. Setup takes one line.
Verify AI Data Transformations in Node.js
Use @latentmachine/verify to check whether AI-generated data transformations are consistent, infer deterministic rules from examples, and apply them to new data. Zero dependencies. Works in any Node.js pipeline, CI job, or automation script.
Audit a CSV Migration With One Paste
Before importing migrated CSV records, verify that every row followed the same transformation rule. Paste original and transformed data into Latentmachine Verify to catch inconsistent rows.
Extract Data from SQL INSERTs by Example
Use Latentmachine to parse SQL INSERT statements from dumps or seed files, turn rows into JSON-compatible data, and reshape them into JSON, CSV, YAML, TOML, or .env output from examples.
Find the Inconsistent Rows in Any Batch
When a batch of transformed records looks mostly right, Latentmachine Verify can identify the exact rows that break the pattern by inferring the majority rule.
How to Verify LLM Data Transformations
LLMs can reshape data quickly, but batch outputs can drift row by row. This guide shows how to verify ChatGPT-style data transformations by checking whether every transformed record follows one deterministic rule.
When AI Gets Your Data Transformation Almost Right
Language models are fast at reshaping data. They are also non-deterministic, quietly inconsistent across rows, and impossible to audit after the fact. This article explains the specific failure modes and what to use when you need every row to be correct.
Build jq Queries by Clicking, Not Typing
Latentmachine's jq Builder generates jq expressions from selected JSON values or a desired output shape. It verifies the query, shows JSONPath equivalents, and asks when your selection is ambiguous. No jq syntax required.
Build a Regex by Showing Examples, Not Writing Syntax
Latentmachine's Regex Builder synthesizes a regular expression from match and reject examples. It verifies the pattern before showing it, explains each token in plain English, supports named captures, and refuses when your examples are ambiguous.
How to Catch Inconsistent Data Transformations Before They Reach Your Database
Latentmachine Verify checks whether transformed records follow one deterministic rule. Paste original and transformed data, and it flags the rows that break the pattern. Works for LLM-generated output, manual conversions, and automated pipeline output.
Convert .env Files to JSON, YAML, or TOML by Example
Use Latentmachine to convert .env files to JSON, YAML, TOML, or CSV — and back — while renaming keys, restructuring flat variables into nested configs, coercing types, and filtering entries. All from examples.
The Complete Guide to Transforming Structured Data by Example
Everything you need to know about transforming JSON, CSV, YAML, TOML, XML, and .env data by showing before-and-after examples instead of writing code. Covers real-world scenarios, cross-format conversion, the diagnosis system, code export, and when to use a different tool.
Transform Structured Data Without Writing Code
Latentmachine infers data transformation rules from before-and-after examples. No scripting, no query language, no syntax. Show the input, show the output, and the engine figures out the rule. Works for JSON, CSV, YAML, TOML, XML, and .env.
Convert XML to JSON by Example
Use Latentmachine to convert XML, RSS, SOAP-like payloads, Maven POMs, and HTML-like fragments into JSON, CSV, YAML, TOML, or .env output while reshaping fields from examples.
Convert TOML to JSON or YAML by Example
Use Latentmachine to convert TOML configs into JSON, YAML, CSV, or .env output while reshaping fields, preserving types, and inspecting the inferred rule.
Export a CLI Tool From Your Transformation Rule
Turn a Latentmachine rule into a standalone command-line tool you can run in scripts, CI pipelines, and cron jobs. One file. No dependencies. No internet. The same guardrails as the browser, printed to your terminal.
How to Map Between Two APIs Without Writing Glue Code
Every integration between two systems needs a transformation layer. Usually you write it by hand. This article shows how to infer the mapping from examples instead — for webhooks, REST APIs, CRM syncs, and payment processors.
How to Transform YAML Configs by Example — Kubernetes, Docker Compose, and CI/CD
Reshape Kubernetes manifests, flatten Docker Compose services, convert YAML configs to JSON or CSV, and clean up CI/CD pipelines by showing before-and-after examples. Handles the Norway problem, anchors, multi-line strings, and nested structures.
Translate Between JSON, CSV, YAML, TOML, XML, and .env — With Structural Transformation
Most format converters change the syntax but not the structure. Latentmachine does both: convert between JSON, CSV, YAML, TOML, XML, and .env while renaming fields, flattening nesting, coercing types, splitting strings, and filtering arrays. All from examples.
How to Convert CSV to JSON (and JSON to CSV) by Example
Convert CSV to JSON, flatten JSON into CSV, and clean tabular exports by showing before-and-after examples instead of writing a parser. Covers headers, type conversion, nested fields, delimiters, and exportable JavaScript.
How to Transform a Shopify Order Webhook into a Clean Record
Reshape Shopify order and webhook payloads into the flat records your database, accounting system, or fulfillment API needs. Map line items, customer fields, money fields, and addresses by example, then export the code for n8n, Make.com, or Node.
From 47 Seconds to 10 Milliseconds: How We Fixed the Inference Engine
Latentmachine's rule inference engine timed out on payloads with more than 10 fields. The fix was not caching, parallelism, or a faster language. It was asking a better question before doing expensive work.
How Latentmachine Infers a Data Transformation Rule From Examples
A walkthrough of the symbolic inference engine behind Latentmachine — how it perceives structure, generates candidates, scores by simplicity, refuses to guess, and earns the right to say 'safe'. No neural networks. No language models. Just search and discipline.
What ARC Made Me Build: A Curiosity-Driven Experiment in Symbolic Reasoning
A late-night curiosity about the primitive structure of intelligence led to an experimental reasoning engine, symbolic data transformations, and a deeper question about who gets to build complex systems now.
Hundreds of Checks That Run Before You See a Result: How Latentmachine Validates Its Own Engine
Latentmachine runs hundreds of checks across rule inference, format translation, YAML handling, export generation, CLI behavior, fixtures, presets, and performance. This article explains how the test system works and why it exists.
From Inferred Rule to Working Code: How Latentmachine Generates JavaScript
Latentmachine infers a symbolic rule, then exports it as runnable JavaScript. This article explains how the rule-to-code translation works, what the generated code looks like, and why it handles edge cases you did not think about.
How an AI Research Puzzle Became a JSON Transformation Tool
Latentmachine started as an experiment with ARC, a reasoning benchmark from AI research. The core idea: figure out rules from examples, not training data. This article explains the origin, what transferred, and why it matters for a tool that reshapes your data.
How Latentmachine Figures Out the Rule: A Plain Explanation of Program Synthesis
Latentmachine infers data transformation rules from examples using program synthesis. This article explains the pipeline step by step: structural diffing, candidate generation, validation, cost-based selection, and diagnosis.
How to Verify Every Claim This Tool Makes
Latentmachine claims zero server calls, deterministic inference, and honest diagnosis. Every one of these claims is verifiable in your browser without access to the source code repository. Here is how to check them yourself.
Latentmachine vs jq, JSONata, and Writing It Yourself: An Honest Comparison
There are many ways to transform structured data. jq, JSONata, lodash, visual mappers, LLMs, and hand-written functions all work. This article compares them honestly, including where Latentmachine is not the right choice.
Nulls, Empty Arrays, and Deeply Nested Nightmares: How the Engine Handles Edge Cases
What happens when you paste ugly JSON into Latentmachine? Deeply nested payloads, null values, empty arrays, mixed types, missing fields. This article walks through the edge cases and what the engine does with each one.
How to Reshape JSON Data by Showing Examples
Learn how to transform JSON payloads, reshape API responses, and clean messy exports by showing before-and-after examples instead of writing scripts. Includes Stripe, WordPress, HubSpot, and Airtable examples with exportable JavaScript.
Reshaping CMS Exports for Migration Without Writing a Script
How to use Latentmachine to transform WordPress, Airtable, and other CMS exports into import-ready JSON for Webflow, Contentful, or any target system. Covers date formatting, tag splitting, field cleanup, and type conversion.
The Tool That Refuses to Guess
Most data transformation tools optimize for producing output. Latentmachine optimizes for telling you when it should not. Here is why a tool that refuses to guess is more useful than one that always has an answer.
Using Latentmachine with n8n and Make.com Workflows
How to use Latentmachine to generate transformation code for n8n Code nodes and Make.com JavaScript modules. Paste a webhook payload, show the shape you need, and export ready-to-use code. Works with JSON, CSV, YAML, TOML, XML, and .env inputs.
What Is Programming by Example (and Why It Keeps Coming Back)
Latentmachine belongs to a field called Programming by Example that has existed since the 1970s. This article explains the idea, traces the lineage from early research through Flash Fill to ARC, and shows where Latentmachine fits.
When Your Transformation Rule Is Wrong (And How to Know Before It Breaks)
Most data transformation tools give you output and hope for the best. Latentmachine gives you a diagnosis first. Learn how confidence assessment, contradiction detection, and ambiguity triage work under the hood.
Why Latentmachine Picks the Simplest Rule (and How the Cost Model Works)
When multiple transformation rules fit the same examples, Latentmachine picks the simplest one. This article explains the cost-based scoring model, why simplicity matters for correctness, and how the engine handles ties.
Why Latentmachine Runs Entirely in Your Browser
Latentmachine processes all data client-side with zero server calls. Here is why that matters for sensitive payloads, and how the architecture works.
You Do Not Need to Write That Transformation Function
A calm but honest look at the time developers spend writing data transformation code that a rule engine could infer in milliseconds. Includes a self-assessment for when to write it yourself and when to let the machine figure it out.