{"title":"Exercise: Prompt Templates and Pipelines","description":null,"content":"# Exercise 1: Chain Two Prompts into a Pipeline\n\nBuild a two-step pipeline where the output of the first prompt becomes the input to the second: summarize a passage, then classify the summary's tone. Each step is its own prompt, and the steps run in sequence.\n\nReal tasks are often too much to ask of a single prompt. Splitting them into a pipeline of focused steps makes each prompt simpler, easier to test, and easier to swap out. The key idea is that the output of one step feeds the next, so you design each prompt around the shape of data it receives and produces.\n\n### Requirements\n\n- Step 1: summarize a passage into one sentence.\n- Step 2: take that summary and classify its tone as `positive`, `negative`, or `neutral`.\n- Pass the first step's output into the second step's prompt.\n- Print the summary and the final tone.\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\nimport os\nfrom dotenv import load_dotenv\n\nload_dotenv()\n\nclient = OpenAI(\n api_key=os.getenv(\"OPENROUTER_API_KEY\"),\n base_url=\"https://openrouter.ai/api/v1\",\n)\n\npassage = (\n \"The new release fixed the long-standing sync bug that frustrated many users, \"\n \"and the team shipped it ahead of schedule. Early feedback has been enthusiastic.\"\n)\n\ndef summarize(text: str) -> str:\n # TODO: prompt the model to summarize text in one sentence and return it\n pass\n\ndef classify_tone(summary: str) -> str:\n # TODO: prompt the model to classify the summary's tone as\n # positive, negative, or neutral, and return just the label\n pass\n\n# TODO: run step 1, feed its output into step 2, and print both\n```\n\n\n### Expected Output\n\nThe summary feeds straight into the tone classifier, and the positive passage is labeled positive.\n\n\n```plaintext\nSummary: The team shipped an early release that fixed a major sync bug to enthusiastic feedback.\nTone: positive\n```\n\n\n### Solution\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\nimport os\nfrom dotenv import load_dotenv\n\nload_dotenv()\n\nclient = OpenAI(\n api_key=os.getenv(\"OPENROUTER_API_KEY\"),\n base_url=\"https://openrouter.ai/api/v1\",\n)\n\npassage = (\n \"The new release fixed the long-standing sync bug that frustrated many users, \"\n \"and the team shipped it ahead of schedule. Early feedback has been enthusiastic.\"\n)\n\ndef summarize(text: str) -> str:\n response = client.chat.completions.create(\n model=\"openai/gpt-5.4-mini\",\n messages=[\n {\"role\": \"system\", \"content\": \"Summarize the text in one sentence.\"},\n {\"role\": \"user\", \"content\": text},\n ],\n )\n return response.choices[0].message.content.strip()\n\ndef classify_tone(summary: str) -> str:\n response = client.chat.completions.create(\n model=\"openai/gpt-5.4-mini\",\n messages=[\n {\n \"role\": \"system\",\n \"content\": \"Classify the tone as positive, negative, or neutral. Reply with only the label.\",\n },\n {\"role\": \"user\", \"content\": summary},\n ],\n )\n return response.choices[0].message.content.strip()\n\nsummary = summarize(passage)\ntone = classify_tone(summary)\n\nprint(f\"Summary: {summary}\")\nprint(f\"Tone: {tone}\")\n```\n\n\nEach step is a self-contained function with its own prompt and a clear input and output type, which is what makes the pipeline composable. The wiring is just passing `summary` from the first call into the second. Structuring work this way pays off as pipelines grow: you can reorder steps, cache intermediate results, or replace one step's prompt without touching the others.\n\n---\n\n# Exercise 2: Add a Guard and Retry to a Pipeline\n\nModels do not always honor constraints on the first try. A guarded pipeline checks the output against a rule and re-runs with a firmer prompt when the check fails. Build a summarizer that enforces a word limit: summarize, verify the length, and retry with a stricter instruction if it is too long.\n\n### Requirements\n\n- Write `summarize(text, max_words, strict=False)` that asks for a summary within a word limit, with a firmer instruction when `strict` is set.\n- Write `within_limit` to check the word count.\n- Summarize once; if the result exceeds the limit, re-run with `strict=True`. Print the final summary and its word count.\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\nimport os\nfrom dotenv import load_dotenv\n\nload_dotenv()\n\nclient = OpenAI(\n api_key=os.getenv(\"OPENROUTER_API_KEY\"),\n base_url=\"https://openrouter.ai/api/v1\",\n)\n\narticle = \"Our new release cuts cold-start latency by 40 percent and adds streaming responses, which testers say makes the app feel far more responsive in everyday use.\"\n\ndef summarize(text, max_words, strict=False):\n # TODO: ask for a summary in at most max_words words.\n # When strict is True, make the instruction firmer about the word limit.\n ...\n\ndef within_limit(summary, max_words):\n return len(summary.split()) <= max_words\n\n# TODO: summarize with max_words=12. If it exceeds the limit, re-run with strict=True.\n# Print the final summary and its word count.\n```\n\n\n### Expected Output\n\nIf the first summary is over the limit, the guard re-runs with a stricter prompt and the final result fits.\n\n\n```plaintext\nNew release cuts cold-start latency 40% and adds streaming for a more responsive app.\nWord count: 12\n```\n\n\n### Solution\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\nimport os\nfrom dotenv import load_dotenv\n\nload_dotenv()\n\nclient = OpenAI(\n api_key=os.getenv(\"OPENROUTER_API_KEY\"),\n base_url=\"https://openrouter.ai/api/v1\",\n)\n\narticle = \"Our new release cuts cold-start latency by 40 percent and adds streaming responses, which testers say makes the app feel far more responsive in everyday use.\"\n\ndef summarize(text, max_words, strict=False):\n instruction = f\"Summarize the text in at most {max_words} words.\"\n if strict:\n instruction += \" This is a hard limit. Count the words and do not exceed it.\"\n response = client.chat.completions.create(\n model=\"openai/gpt-5.4-mini\",\n messages=[{\"role\": \"system\", \"content\": instruction}, {\"role\": \"user\", \"content\": text}],\n )\n return response.choices[0].message.content.strip()\n\ndef within_limit(summary, max_words):\n return len(summary.split()) <= max_words\n\nMAX_WORDS = 12\nsummary = summarize(article, MAX_WORDS)\nif not within_limit(summary, MAX_WORDS):\n summary = summarize(article, MAX_WORDS, strict=True)\n\nprint(summary)\nprint(\"Word count:\", len(summary.split()))\n```\n\n\nThe guard turns a soft request (\"at most 12 words\") into an enforced one by checking the output and acting on the result. Re-running with a stricter instruction is the cheapest correction; a production version might also lower `max_tokens` or fall back to truncation after a couple of tries. This check-and-retry shape generalizes to any constraint you can verify in code, such as valid JSON, a required field, or a banned word, and it is what makes a pipeline reliable rather than hopeful.","pageType":"ai-engineering"}

Get Premium