Do-it-yourself game plans

What should we build first?

Frontier Development research deskSection 02 · Page 1 of 15Last updated 26 October 2026

Build a repeatable test and a cost ledger before a trained model. The first milestone is an honest baseline and a decision rule that can reject adaptation.

This plan is for a software team with an accountable product owner and access to an engineer who can maintain the eventual service. It covers extraction and support workflows. Timings below are editorial planning allowances, not observed industry averages or supplier promises; data-rights approval, expert review and infrastructure access can dominate the schedule.

TRACE THE EVIDENCE

Section source library

This page contains no direct external citations. These are the section’s references; they do not independently validate the desk’s proposed scoring or methods.

Demystifying evals for AI agents — Anthropic, 9 January 2026 →www.anthropic.com
Quantization — Hugging Face PEFT documentation →huggingface.co
LoRA Learns Less and Forgets Less — Biderman et al., revision 20 September 2024 →arxiv.org
Pricing — Together AI, snapshot checked 26 October 2026 →www.together.ai
GroupKFold — scikit-learn documentation →scikit-learn.org
A Field Guide to Rapidly Improving AI Products — Hamel Husain →hamel.dev
Guidance on HIPAA & Cloud Computing — HHS Office for Civil Rights →www.hhs.gov
TimeSeriesSplit — scikit-learn documentation →scikit-learn.org
Extend — Extraction Overview →docs.extend.ai
SFT Trainer — Hugging Face TRL documentation →huggingface.co
RouteLLM: Learning to Route LLMs with Preference Data — Ong et al., revision 23 February 2025 →arxiv.org