AI Research Loop Template

A lightweight framework for running small AI experiments with asymmetric upside.
Built for people who want clarity without overcomplication.

---

Why

Most AI experiments fail because they’re too big, too vague, or never written down.
This template forces you to define:

• Hypothesis (what you believe will happen)
• Setup (tools/models you’ll use)
• Steps (≤3)
• Result (pass/fail + notes)
• Next (iterate / kill / scale)

It works like a barbell strategy: tiny capped downside, open-ended upside.

---

How to Use

1. Clone/download this repo.
2. Copy the ai_experiment_template.md file.
3. Fill it in before you start any new AI test.
4. Keep each experiment to ≤60 minutes or ≤$20 cost.
5. Track results over time → patterns emerge.

---

Example

# AI Experiment
Hypothesis: If I use GPT-4 to summarize call transcripts, I’ll save 30 min per meeting.
Setup: GPT-4 API + Zoom transcript
Steps:
1. Export transcript
2. Prompt GPT-4 with "3 key decisions, 2 risks"
3. Paste result into meeting notes
Result: Saved 25 min, but missed 1 risk.
Next: Refine prompt to include "financial risk".