Lattice

A lattice of prompts for serious work

Curated prompts for literature reviews, Aion workflows, datasets, writing, and evaluation. Copy once, adapt for your model.

10 prompts

Researchpapersoutline

Literature review outline

Structure a focused lit review for an ML or scientific topic.

You are a research assistant. Create a literature review outline for the topic: {{TOPIC}}.

Requirements:
- Group papers by research question, not by year alone
- For each group, list 3–5 representative works with one-line contribution + limitation
- Call out open problems and evaluation gaps
- Suggest 5 search keywords and 3 conference/journal venues
- Keep the tone precise and evidence-first; avoid hype

Paste into any LLM

Researchexperimentsscientific method

Hypothesis and experiment plan

Turn a vague idea into a testable hypothesis and minimal experiment.

Help me turn this idea into a testable research plan:

Idea: {{IDEA}}

Deliver:
1. A single clear hypothesis (falsifiable)
2. Variables: independent, dependent, controls
3. Minimal experiment design (data, baseline, metrics)
4. Failure modes and what would refute the claim
5. A one-paragraph write-up suitable for a lab notebook

Paste into any LLM

CodingAionPython

Aion helper script

Ask for a small Python script that uses Aion-style research utilities.

Write a concise Python script for this task using clear research-engineering style (as if building on Aqwel Aion primitives):

Task: {{TASK}}

Constraints:
- Prefer readable functions over clever one-liners
- Include a `if __name__ == "__main__":` entrypoint
- Print key metrics or outputs
- Add brief comments only where non-obvious
- Note any Aion modules I should check in the docs

Paste into any LLM

CodingdebuggingML

Debug ML pipeline

Systematic debugging checklist for training or evaluation failures.

I have a failing ML pipeline. Debug systematically.

Symptom: {{SYMPTOM}}
Stack/context: {{CONTEXT}}

Please:
1. List the top 5 most likely root causes ranked by probability
2. Give a concrete check for each (command, assert, or plot)
3. Propose a minimal fix for the most likely cause
4. Tell me what log lines or shapes to print next

Paste into any LLM

DatasetsschemaCSV

Dataset schema designer

Design columns, types, and documentation for a research dataset.

Design a clean dataset schema for:

Use case: {{USE_CASE}}

Provide:
- Column name, type, allowed values / ranges, missingness policy
- 3 example rows (realistic)
- Suggested filename and folder layout
- License and citation stub
- Potential leakage risks and how to avoid them

Paste into any LLM

Datasetssynthetictesting

Synthetic data brief

Specify synthetic data generation for testing models safely.

Draft a synthetic data generation brief for:

Goal: {{GOAL}}

Include:
- Distributions and correlations to simulate
- Noise / missingness models
- Train/val/test split policy
- Sanity checks (stats + plots)
- Explicit statement of what the data does NOT represent in the real world

Paste into any LLM

Writingabstractpaper

Technical abstract

Draft a tight abstract for a paper or lab note.

Write a technical abstract (150–200 words) for:

Title: {{TITLE}}
Contribution: {{CONTRIBUTION}}
Method: {{METHOD}}
Results: {{RESULTS}}

Style: clear, no marketing language, past tense for completed work, mention limitations in one clause.

Paste into any LLM

WritingREADMEdocs

Open-source README

Generate a README structure for a research library or tool.

Create a README.md outline and fill each section for this project:

Name: {{NAME}}
One-liner: {{ONE_LINER}}
Audience: {{AUDIENCE}}

Sections: Overview, Install, Quickstart, Features, Docs links, Contributing, License, Citation.
Keep install commands copy-pasteable.

Paste into any LLM

EvaluationLLMrubric

LLM evaluation rubric

Build a scoring rubric for qualitative LLM outputs.

Create an evaluation rubric for LLM answers on this task:

Task: {{TASK}}

Deliver:
- 4–6 criteria with 1–5 scale descriptors
- Pass / fail threshold
- Example of a high-scoring and low-scoring response
- How to reduce judge bias when using an LLM-as-judge

Paste into any LLM

Evaluationablationexperiments

Ablation study planner

Plan ablations that isolate what actually drives results.

Design an ablation study for:

Claim: {{CLAIM}}
System components: {{COMPONENTS}}

Provide:
- Which components to remove/replace first
- Metrics and statistical checks
- Table template for results
- What would convince a skeptical reviewer

Paste into any LLM