Free ML Engineering Prompts for ChatGPT

Act as a senior ML engineer with 10+ years specializing in ML engineering for data and product teams. I will give you the brief; you will deliver a ship-ready ML plan.

Brief: [PASTE BRIEF HERE].
Constraints: must be specific, measurable, and grounded in ML engineering best practice. Avoid generic advice and obvious tips.

Deliver:
1. The full ML plan (the actual artifact, not a description of it).
2. Three sharpening notes - what you would test or improve first.
3. One contrarian angle most ML engineers miss.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

You are a brutal but constructive ML engineer reviewing my ML plan. Your job is to make it 2x better, not to be polite.

My ML plan: [PASTE HERE].
Target model quality: [STATE TARGET].
Audience: data and product teams.

Return:
- 5 specific weaknesses, each tied to model quality.
- A rewritten ML plan that fixes them.
- A diff-style explanation of what changed and why.

Be blunt. Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Generate 10 meaningfully different ML plans from the same brief. Each variant must hit a different angle - not paraphrases.

Brief: [PASTE].
Audience: data and product teams.

For each variant provide:
- Angle name (1-3 words).
- Hook / opening line.
- Full ML plan.
- The single psychological lever it pulls (loss aversion, status, novelty, etc.).

End with your top pick and a one-line reason. Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Pick the single best-known ML engineering framework for this situation, name it, then walk me through applying it to my brief step by step.

My situation: [PASTE].
Goal: model quality.

Output:
1. Framework name + 1-line origin (so I can verify).
2. Each step labelled, with my inputs filled in.
3. The resulting ML plan.
4. Where the framework breaks down - and what to swap in.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Take my existing ML plan and rewrite it cleanly for three distinct audiences. Keep the core promise; change the vocabulary, references, and emotional register.

My ML plan: [PASTE].

Audiences:
A) data and product teams (current).
B) A skeptic who has been burned before.
C) An expert peer who could spot fluff in two seconds.

For each: full rewrite + 2-line note on what shifted. Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Compress the strongest possible ML plan into exactly 50 words. Every word must earn its place.

Brief: [PASTE].
Audience: data and product teams.

Deliver:
- The 50-word ML plan.
- The 3 words you would protect if forced to cut to 30.
- The cheap word you almost used and why you killed it.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Produce a ML plan that is optimized to be found and shared in the ML engineering space, not just to read well.

Topic: [PASTE].
Audience: data and product teams.
Primary keyword/phrase: [PASTE].

Deliver:
- The ML plan, with the primary phrase used naturally in title, opener, and one mid-point anchor.
- 5 semantic keywords you wove in (and where).
- 3 share-bait one-liners I could pull as social hooks.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Build a comparison matrix that ranks options for my ML engineering decision honestly.

Options: [LIST 3-6].
My priority: model quality.
Constraints: [PASTE].

Deliver:
1. A table - options × criteria - scored 1-5 with a one-line justification per cell.
2. The weighted winner.
3. The "wrong but obvious" pick most ML engineers would default to, and why it loses.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Embody a precise data and product teams persona and react to my ML plan as they would, out loud.

The persona: [PASTE 3-5 traits - role, fear, current solution, last frustration].
My ML plan: [PASTE].

Deliver:
- 3 internal-monologue paragraphs as the persona reading the ML plan.
- The exact line where they would close the tab - and why.
- 2 edits that would make them keep reading.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Enumerate the edge cases and failure modes that could break my ML plan in production / in market / in front of data and product teams.

My ML plan: [PASTE].
Context: ML engineering.

Deliver:
- 12 edge cases, ranked by likelihood × damage.
- For each: the trigger, the symptom, and the cheapest mitigation.
- The single edge case I should design around first.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Stress-test my ML engineering plan and surface what could go wrong, with mitigations.

Plan: [PASTE].
Stakes: model quality.

Deliver:
1. 7 risks across execution, market, technical, legal, reputational.
2. For each - probability (L/M/H), impact (L/M/H), and a mitigation that costs less than the worst case.
3. The 1 risk worth accepting and the 1 risk worth killing the plan over.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Convert a great ML plan into a reusable template I can fill in repeatedly.

Reference ML plan: [PASTE].
What stays fixed: the structure and rhythm.
What varies: the inputs.

Deliver:
- The template with clearly marked [VARIABLES].
- A one-line description of each variable and example values.
- 2 worked examples using different inputs.
- The 1 line I should never let a junior change.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Design a 30-day plan to take me from where I am now to shipping a credible ML plan in ML engineering.

Starting point: [PASTE].
Time available per day: [PASTE].
End state: model quality.

Deliver:
- Week 1-4 milestones (1 sentence each).
- Daily 30-minute focus for every day, grouped by week.
- The 3 things I should NOT do during these 30 days.
- The checkpoint that proves I'm on track at day 14.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

I am seeing symptoms in my ML engineering work. Diagnose the most likely root causes and propose tests to confirm.

Symptoms: [LIST 3-6].
What I have already ruled out: [PASTE].
Tools available: [PASTE].

Deliver:
1. 3 candidate root causes, ranked by likelihood with a 1-line reason.
2. The fastest test to disprove each.
3. The order to run those tests, and stop conditions.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Design the end-to-end workflow a ML engineer would run to produce a high-quality ML plan repeatedly.

Volume target: [PASTE].
Team size: [PASTE].
Quality bar: model quality.

Deliver:
- The workflow as a numbered sequence of steps.
- For each step: input, output, owner, tool, and time-box.
- Where to insert review gates without slowing the pipeline.
- The bottleneck step and how to relieve it.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Design a measurement plan for model quality in this ML engineering context.

Goal: model quality.
Audience for the dashboard: [PASTE].
Available data sources: [PASTE].

Deliver:
- The 1 north-star metric - defined precisely.
- 3 input metrics that move it, with formulas.
- 3 guardrail metrics so we don't optimize the wrong thing.
- Dashboard layout sketch (sections, charts, refresh cadence).
- The 1 vanity metric I am tempted to track and should not.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Build an objection-handling script for my ML plan aimed at data and product teams.

My offer / position: [PASTE].
The 3 most common objections I hear: [PASTE].

Deliver:
- For each objection: validate, reframe, evidence, ask.
- 2 objections I am probably not hearing but should expect.
- The single phrase to never say in response, and why.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Design an A/B test for my ML plan with one clear hypothesis.

Current ML plan: [PASTE].
Hypothesis (or what I'm curious about): [PASTE].
Traffic / sample size available: [PASTE].

Deliver:
- The hypothesis sharpened to one sentence.
- Variant A vs Variant B - only one variable changed.
- Primary metric and minimum detectable effect.
- Test duration and stop conditions.
- The decision rule before I peek at results.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Storyboard my ML plan as a 7-beat narrative arc.

Subject: [PASTE].
Audience: data and product teams.
Emotional outcome I want: [PASTE].

Beats: Hook → Stakes → Conflict → Attempt → Setback → Insight → Resolution.

For each beat: 1 sentence of action + 1 sentence of feeling. End with the single image the audience walks away with. Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Build a glossary of the 20 most important terms in ML engineering as it relates to ML plans and data and product teams.

For each term:
- The term.
- A precise 1-sentence definition (no jargon recursion).
- 1 concrete example.
- The most common misuse I should watch out for.

End with the 1 term that is overused and meaningless. Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Produce a one-page cheat sheet for shipping a ML plan in ML engineering.

Audience: data and product teams.
Bias toward action, not theory.

Sections:
1. The 5-step quick path.
2. 3 hard rules (never break).
3. 3 soft rules (break with a reason).
4. Top mistake at each step.
5. The single check before publishing / shipping / sending.

Plain text, dense, under 400 words. Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

List the most damaging anti-patterns in ML engineering when producing a ML plan.

Deliver:
- 8 anti-patterns.
- For each: a named label, a 1-line description, a real-sounding example of the failure, and the corrective principle.
- The anti-pattern that looks like best practice from the outside.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

My ML plan or ML engineering effort went wrong. Build me a recovery plan.

What happened: [PASTE].
Who noticed: [PASTE].
Reversibility (1=easy, 5=baked-in): [PASTE].

Deliver:
- The first 24 hours: communications and operational steps.
- The next 7 days: trust-rebuild moves.
- The 30-day move that turns this into a credibility gain.
- The 1 thing I must NOT do in the first 24 hours.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Build a 90-day roadmap for serious progress on model quality in ML engineering.

Current state: [PASTE].
End state: [PASTE].
Resources: [PASTE].

Deliver:
- Month 1 / Month 2 / Month 3 themes (1 line each).
- 3-5 outcomes per month - each measurable.
- Dependencies and the order they must clear.
- The single bet I should kill if month 1 underdelivers.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.

Build me a swipe file of exceptional ML plans in ML engineering to study, not copy.

My focus: model quality.
Audience I serve: data and product teams.

Deliver:
- 10 exemplary ML plans (real or plausibly real) with 1-line context for each.
- For each - the one technique to steal and the one tic to avoid.
- 3 patterns that show up across most of them.
- The exemplary ML plan that is overrated and why.

Ask one clarifying question only if a hard blocker remains; otherwise proceed with stated assumptions. Use plain language, no fluff, no filler. Quote evidence when citing sources.