DeepSeek Prompts

Act as my patient tutor. Solve this problem step by step: [PROBLEM, e.g. a calculus optimization question or a physics word problem]. Use DeepThink reasoning and show every step: restate what is given and what is asked, choose the method and say why ([METHOD] because [REASON]), work through the algebra with each transformation on its own line, box the final answer, then verify it by [VERIFICATION METHOD, e.g. plugging the answer back in or checking units]. After the solution, give me [NUMBER] similar practice problems at the same difficulty without solutions, and list the [NUMBER] most common mistakes students make on this problem type. If the problem as stated is ambiguous, ask me [NUMBER] clarifying questions before solving.

You are my strategy analyst. I run [MY BUSINESS TYPE] in [MARKET OR REGION], and my main competitors are [COMPETITOR LIST]. Think through this deeply and deliver: 1) a comparison table on [CRITERIA, e.g. pricing, features, reputation, marketing], 2) each competitor's likely strategy based on [EVIDENCE YOU HAVE], 3) [NUMBER] gaps in the market none of them serve well, 4) [NUMBER] moves I could make in the next [TIMEFRAME], ranked by expected impact versus effort, 5) the single biggest threat to my business and a mitigation plan. Reason explicitly about your assumptions and mark each claim as [FROM MY INPUT], [REASONED], or [NEEDS VERIFICATION]. End with the [NUMBER] questions I should research next.

You are my senior engineer. I will paste [CODE DESCRIPTION, e.g. a Python API endpoint] of about [LINE COUNT] lines. Review it with deep reasoning: 1) what the code does, in [NUMBER] sentences, 2) [NUMBER] bugs or edge cases, each with the exact line, why it breaks, and a fixed version, 3) [NUMBER] performance or security concerns ranked by severity, 4) whether the architecture fits [PROJECT CONTEXT, e.g. a high-traffic REST API] and what you would change, 5) a rewritten version of the weakest function. Think through the execution path explicitly before concluding. Do not invent behavior of libraries I did not mention; ask if you need to see [MISSING CONTEXT].

You are my document analyst. I will upload [FILE TYPE, e.g. a PDF annual report or a text contract] about [TOPIC]. Once uploaded: 1) confirm the document type, page count, and date, and note any pages that are unreadable, 2) extract [DATA POINTS, e.g. all financial figures or all deadlines and obligations] into a clean table, 3) summarize the [NUMBER] most important sections in [WORD COUNT] words total, 4) flag [NUMBER] items that need a human expert ([LEGAL, FINANCIAL, OR TECHNICAL] review), 5) list questions the document raises but does not answer. Quote exact wording, with page numbers, for anything I might need to cite to [STAKEHOLDER]. If the upload has OCR errors, tell me which parts are uncertain.

You are my backend architect. I want to add AI features ([FEATURES, e.g. support-ticket classification and reply drafting]) to my [APP TYPE] using the DeepSeek API because of its low cost. Design the integration: 1) which model for which task (V3 for [TASK A], R1 for [TASK B]) and why, 2) the [NUMBER] API endpoints I need with request and response shapes, 3) prompt templates for each feature with [VARIABLES] marked, 4) a cost estimate ([REQUESTS PER DAY] requests times average [TOKENS] tokens) with the math shown, 5) error handling and fallbacks for rate limits, timeouts, and bad outputs, 6) [NUMBER] tests to run before launch. Keep the design simple enough for a [TEAM SIZE]-person team to build in [TIMEFRAME]. Flag anything that would genuinely need a more expensive model instead.

Act as my infrastructure advisor. I am considering self-hosting an open-weights DeepSeek model for [USE CASE, e.g. internal document Q and A] because of [REASON, e.g. data privacy]. Evaluate it: 1) which model size fits [HARDWARE, e.g. 2x RTX 4090 or a cloud VM with X GB RAM], 2) expected performance (tokens per second, quality trade-offs versus the hosted API), 3) the [NUMBER] setup steps in order (quantization, serving framework, API compatibility), 4) total cost comparison of self-hosted versus API over [TIMEFRAME] at [USAGE LEVEL], with the math shown, 5) [NUMBER] risks (updates, security, quality drift) and mitigations. Be honest about when self-hosting is NOT worth it. End with a clear recommendation for my situation.

You are my market researcher. I am exploring [BUSINESS IDEA] for [TARGET CUSTOMERS] in [REGION]. Search the web and report: 1) market size estimates with sources and dates ([NUMBER] sources minimum), 2) [NUMBER] existing competitors and what each does differently, 3) [NUMBER] customer pain points quoted from reviews or forums, with sources, 4) pricing benchmarks in [CURRENCY], 5) [NUMBER] reasons this idea might fail, stated bluntly. Mark every number with its source and date; mark anything you could not verify as [UNVERIFIED LABEL, e.g. unverified]. End with your honest assessment (proceed, pivot, or drop) and the single strongest reason.

You are my decision coach. I am stuck choosing between [OPTION A] and [OPTION B] for [SITUATION]. Build me a reusable framework: 1) the [NUMBER] criteria that actually matter for this decision (not generic ones), with weights summing to 100 based on my stated priorities ([PRIORITIES]), 2) a scoring table where I fill in 1 to 10 per criterion, 3) the [NUMBER] hidden factors people usually miss in decisions like this, 4) a regret test: which choice I would regret more in [TIMEFRAME] if it went wrong, and why, 5) a one-paragraph pre-mortem for each option (assume it failed; what went wrong). Reason through the weights explicitly and challenge any priority of mine that seems inconsistent with [MY STATED GOAL].

You are my prompt engineer. I need prompts for [TASK, e.g. extracting structured data from customer emails] that I will run on [MODEL CHOICE: DeepSeek V3 for speed, or R1 for reasoning]. Write: 1) the system prompt (role, context, output format as [FORMAT, e.g. JSON with these keys]), 2) the user prompt template with [VARIABLES], 3) [NUMBER] few-shot examples (input to ideal output), 4) edge-case handling rules for [EDGE CASES], 5) a test set of [NUMBER] tricky inputs I should try, with expected outputs. Explain which parts rely on R1's reasoning versus V3's speed, and where to put the constraint if outputs drift ([CONSTRAINT PLACEMENT]). Keep prompts robust to messy real-world inputs in [DOMAIN].

You are my teacher. Explain [COMPLEX TOPIC, e.g. how neural networks learn] to me. I am [BACKGROUND, e.g. a marketing manager with no math background] and I need to understand it well enough to [GOAL, e.g. explain it to my team]. Teach it in layers: 1) a one-paragraph intuition with an analogy from [FAMILIAR DOMAIN], 2) the [NUMBER] key ideas, each with a concrete example, 3) the reasoning chain connecting them (show how idea 1 leads to idea 2), 4) [NUMBER] common misconceptions and why each is wrong, 5) a 5-question self-test with answers hidden. Check my understanding: after layer 2, ask me [NUMBER] questions before continuing. Use plain language; introduce a technical term only when I need it, and define it immediately.