Research & Engineering Notes

Insights on LLMs, training data, evaluation, and agentic AI from the Calibrion team.

Synthetic Data, Revisited: What Remained the Same and What Changed

Eighteen months ago we wrote about the upcoming data crisis. We also wrote that LLM-generated training data wasn't the answer. But the synthetic data boom of 2025–26 actually proved: that synthetic data generation is doable and cheap and with a verification step, it can work just as good for many use cases.

Synthetic dataJul 7, 2026M. Farahmand

Sustaining the Future of LLMs: Overcoming the Data Challenge

Human-generated web data will run dry in the next few years, and using LLMs to generate their own training data has proven a dead end. Sustaining the future of LLMs means rethinking the tools we use to create training data.

LLMsJan 21, 2025M. Farahmand

Fine-tuned LLMs: A Smart Choice on Companies' AI Roadmap

Companies are navigating the ever-changing landscape of AI with a growing need for efficient and cost-effective solutions. We focus on LLMs and explore their unique advantages compared to traditional machine learning methods.

LLMsDec 19, 2024M. Farahmand

Tell us about your data

If you already have a dataset, tell us what it contains, how it is used, and what is not working.

If you need a new dataset, describe the task and what the data must cover.

Useful details:

  • Dataset type and approximate size
  • Training or evaluation task
  • Defects, missing coverage, or target outcome