research-prompts

A small collection of procedure-as-prompt templates I use to integrate AI tools into a computer engineering research workflow: reading papers faster, tightening drafts near deadlines, preserving writing voice, and reducing hallucinations via cross-checking.

Most prompts are written as repeatable procedures (clear inputs, strict outputs, guardrails, and a verification pass), so you can reuse them across projects and teach them to students. For best results, enable your model’s thinking mode (for example, Extended Thinking in ChatGPT or Pro/Thinking in Gemini) when available.

What’s in this repo

• brutal-review/ — This prompt is a near-submission quality gate for research papers. It is designed for the last few hours (or last day) before a deadline, when you need high-signal triage: what must be fixed to avoid rejection or credibility loss, vs what can be polished if time remains.
• chalk-talk/ — This prompt generates a chalkboard-style “chalk talk” slide for teaching a concept: bubbles/circles connected by arrows, short handwritten-style text, and a visual flow that mirrors how an instructor would explain the idea at a board.
• email-scanner/ — This prompt is a daily inbox triage procedure that uses Gemini’s Gmail assistant to surface the few emails you truly need to read from a high-volume inbox. It performs a multi-pass scan over a fixed time window (default: the last two weeks), filters out already-handled threads via a +PROCESSED+ label, then groups the remaining candidates into priority buckets (visits, talks, reviews, letters, overdue, other). Gemini-only (runs in Gmail’s Gemini assistant panel).
• fast-fail/ — A fail-fast ideation workflow for research and engineering: generate novel candidate ideas, then design the cheapest decisive tests (hand-coded micro-tests, intrinsics/ASM, clever FPGA prototypes, minimal sims) to reach high-confidence go/no-go quickly.
• hallucination-detector/ — This prompt is a second-opinion “hallucination detector” for AI-generated content. It scans a draft for five common failure modes and reports issues in order of urgency, with evidence and concrete fixes. Best run on a different vendor/model than the one that produced the draft.
• heilmeier-extractor/ — This prompt is a paper-reading accelerator: it turns a typeset research paper into a structured, reviewer-ready analysis organized around the Heilmeier Catechism (DARPA-style questions that expose the problem, novelty, evidence, and adoption risks).
• orphan-finder/ — This prompt helps you trim a paper without deleting content by finding and fixing paragraph orphans (paragraphs whose last line contains only a few words, wasting vertical space). It identifies candidate paragraphs and proposes minimal, meaning-preserving rewrites that “pull words up” to reclaim full lines and often recover 0.1–0.5 pages near a submission deadline.
• skeleton-prompt/ — This directory contains a generic “procedure-as-prompt” skeleton you can reuse for research and other high-stakes work. The goal is to turn an LLM request into a repeatable procedure with clear inputs, outputs, guardrails, and a built-in verification loop. Use this format for new contributions.
• tough-crowd/ — This prompt is a brutal final-pass reviewer for near-final technical presentations. It prioritizes narrative clarity, cognitive load reduction, and visual-verbal alignment, then outputs a surgical fix list (must-fix vs nice-to-have) to improve speaker-readiness right before delivery.
• writing-voice/ — This package contains two prompts that help an LLM write in your recognizable writing voice: Includes voice-analyzer.prompt and write-in-voice.prompt.

Getting started

1. Pick a directory above and open its README.md.
2. Copy the corresponding .prompt file into your LLM tool.
3. Fill in the SETTINGS block and run on your input (PDF, draft text, inbox, etc.).
4. Treat the output as draft guidance: you remain responsible for correctness, citations, and policy compliance.

Contributions welcome

If you have a prompt that improves research productivity (paper intake, experiment planning, review writing, proposal editing, teaching diagrams, etc.), please open a PR.

A few requests:

• Please format new prompts using the skeleton-prompt/ procedure template (inputs → output contract → guardrails → verification → deliverable).
• Include a short README.md in the new directory explaining when to use it, inputs/outputs, and any vendor/tool requirements (Gemini-only, PDF required, etc.).
• If you are adapting someone else’s idea, attribute the source in the README.

License

Distributed under the Apache 2.0 license, please see LICENSE for details.