Use-case stack · technical leads

Model evals & bake-offs

Compare models on your prompts before locking a stack for the lab.

Same prompts, same temperature notes, pass/fail rubric. OpenRouter/local runners help; spreadsheets keep score.

Preferred / primary

Web LLM / chat

One API key to many model providers — useful for comparing models without N dashboards.

In this stack: Hit many models with one key for bake-offs.

Details Site

Hybrid Research

Spreadsheet as system of record; LLM helps formulas, cleanup, and explanations.

In this stack: Scorecard: prompt, model, pass/fail, notes.

Details Site

Self-hosted LLM / chat

Simplest path to run open models locally (CLI + local API).

In this stack: Local baseline models.

Details Site

App / desktop LLM / chat

Desktop app to download and run local GGUF models with a chat UI.

In this stack: Desktop local comparisons.

Details Site

Hybrid Research

Serious data analysis: notebooks for truth, LLM for boilerplate and explanation.

In this stack: Scripted eval harnesses.

Details Site

Web LLM / chat

Anthropic models strong at long-context writing, code review, and careful analysis.

In this stack: Help design the rubric (not grade itself unchecked).

Details Site

Stacks are editorial starting points — swap tools when policy, budget, or hardware demands it.