Curlscape logo

Blog

Notes from the Curlscape team on building production AI: agents, LLMs, evaluation, and applying AI to engineering and industrial problems.
Surrogate models: where millisecond predictions can be trusted

August 4, 2026

Surrogate models: where millisecond predictions can be trusted

A surrogate trained on a few hundred solver runs predicts your quantity of interest in milliseconds. It is reliable inside the envelope you sampled, and not outside it.

Read more
Automation without verification just ships wrong answers faster

July 28, 2026

Automation without verification just ships wrong answers faster

A sign-convention error flipped a force-and-torque result in an automated simulation study. A verification agent caught it before the report shipped.

Read more
Mistral API Pricing 2026: The Overlooked Budget Powerhouse

July 21, 2026

Mistral API Pricing 2026: The Overlooked Budget Powerhouse

July 2026 Mistral API pricing for every model — text, reasoning, code, vision, voice, and OCR — up to ~90% cheaper than OpenAI and Anthropic. See the breakdown.

Read more
Google Gemini API Pricing Guide 2026: Flash, Pro, and Vertex AI

July 13, 2026

Google Gemini API Pricing Guide 2026: Flash, Pro, and Vertex AI

Current Google Gemini API pricing for 2026: Gemini 3 generation (3.1 Pro, 3.5 Flash, 3.1 Flash-Lite), what changed since 2.5, image generation with Nano Banana, and how Vertex AI costs compare.

Read more
Anthropic Claude API Pricing Guide 2026: Opus, Sonnet, and Haiku Compared

March 11, 2026

Anthropic Claude API Pricing Guide 2026: Opus, Sonnet, and Haiku Compared

Complete Anthropic Claude API pricing for March 2026. Compare Opus, Sonnet 4.6, and Haiku 4.5 with batch discounts, prompt caching savings, rate limits, and real-world cost breakdowns.

Read more
Fine-tuning open models in the real world: Unsloth, Axolotl, and the case for Docker

March 3, 2026

Fine-tuning open models in the real world: Unsloth, Axolotl, and the case for Docker

Production lessons from fine-tuning open models and why Curlscape uses Docker to ensure GPU training environments are reproducible and reliable.

Read more
OpenAI API Pricing Guide 2026: Every Model Compared

February 18, 2026

OpenAI API Pricing Guide 2026: Every Model Compared

Every OpenAI API model priced and compared for 2026, from GPT-5.2 to o4 Mini. Includes real-world cost calculations for chatbots, pipelines, and more.

Read more
The Hidden Math of Text: A Guide to Quantitative Analysis

December 1, 2025

The Hidden Math of Text: A Guide to Quantitative Analysis

Analyze text quantitatively without LLMs — character distributions, structural patterns, and information density: the measurable statistical story in every document.

Read more
Fine-Tune an LLM to Mask PII in 2 Hours with Axolotl — Step-by-Step Tutorial

October 8, 2025

Fine-Tune an LLM to Mask PII in 2 Hours with Axolotl — Step-by-Step Tutorial

Learn to fine-tune an LLM for PII redaction using Axolotl and Modal. Step-by-step tutorial covering QLoRA, dataset preparation, and production deployment for GDPR compliance.

Read more
OpenAI vs Gemini vs Claude: Complete API Pricing Comparison (2026)

September 9, 2025

OpenAI vs Gemini vs Claude: Complete API Pricing Comparison (2026)

Compare API costs for OpenAI GPT-5, Google Gemini, and open-source models. See real pricing per 1M tokens, cost breakdowns for 1M document inference, and when self-hosting saves 5x.

Read more
Full-Text Search at Scale: PostgreSQL vs Elasticsearch vs Vector Search (2026)

May 28, 2025

Full-Text Search at Scale: PostgreSQL vs Elasticsearch vs Vector Search (2026)

Compare full-text search solutions for large datasets: PostgreSQL, Elasticsearch, DuckDB, and vector search. Benchmarks on 3.8M rows with BM25, query latency comparisons, and production implementation tips.

Read more
How to create a MCP Server - An Introduction using SEC EDGAR API

April 14, 2025

How to create a MCP Server - An Introduction using SEC EDGAR API

Step-by-step tutorial on building an MCP server using the SEC EDGAR full-text search API. Learn how to expose tools to LLMs and build AI agents with the Model Context Protocol.

Read more
Book a free consultation