{"title":"Exercise: RAG with Citations and Grounding","description":null,"content":"# Exercise 1: Answer with Inline Citations\n\nGiven a question and a set of numbered context chunks, have the model answer the question and cite the chunk numbers it used inline, so every claim can be traced back to its source.\n\nCitations are what make a RAG answer auditable. Without them, a user has no way to tell which parts of an answer came from the documents and which the model produced on its own. Numbering the chunks and asking the model to reference those numbers turns the answer into something a reader can verify against the original sources.\n\n### Requirements\n\n- Number the context chunks `[1]`, `[2]`, `[3]`, and so on.\n- Instruct the model to answer using only the chunks and to cite the relevant chunk number in brackets after each claim.\n- Print the answer with its inline citations.\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI() # Uses OPENAI_API_KEY from environment\n\nchunks = [\n \"The Python GIL allows only one thread to execute Python bytecode at a time.\",\n \"For CPU-bound work, the multiprocessing module sidesteps the GIL by using separate processes.\",\n \"For IO-bound work, threads still help because the GIL is released during blocking IO.\",\n]\n\nquestion = \"How should I parallelize CPU-bound work in Python given the GIL?\"\n\n# TODO: build a context string with each chunk numbered [1], [2], [3]\n\n# TODO: prompt the model to answer using only the chunks and to cite\n# the chunk number in brackets after each claim it makes\n\n# TODO: print the answer\n```\n\n\n### Expected Output\n\nThe answer reads naturally and carries bracketed citations that point at the chunks each claim came from.\n\n\n```plaintext\nBecause the GIL lets only one thread run Python bytecode at a time [1], CPU-bound work does not speed up with threads. Use the multiprocessing module, which sidesteps the GIL with separate processes [2].\n```\n\n\n### Solution\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI()\n\nchunks = [\n \"The Python GIL allows only one thread to execute Python bytecode at a time.\",\n \"For CPU-bound work, the multiprocessing module sidesteps the GIL by using separate processes.\",\n \"For IO-bound work, threads still help because the GIL is released during blocking IO.\",\n]\n\nquestion = \"How should I parallelize CPU-bound work in Python given the GIL?\"\n\ncontext = \"\\n\".join(f\"[{i + 1}] {chunk}\" for i, chunk in enumerate(chunks))\n\nresponse = client.chat.completions.create(\n model=\"gpt-5.4-mini\",\n messages=[\n {\n \"role\": \"system\",\n \"content\": (\n \"Answer the question using only the numbered context chunks. \"\n \"After each claim, cite the chunk number it came from in brackets, \"\n \"like [1]. Do not use any information outside the chunks.\"\n ),\n },\n {\"role\": \"user\", \"content\": f\"Context:\\n{context}\\n\\nQuestion: {question}\"},\n ],\n)\n\nprint(response.choices[0].message.content)\n```\n\n\nNumbering the chunks gives the model a stable label to point at, and the system prompt makes the citation a required part of the format rather than an afterthought. The result is a grounded answer where each claim names its source, so a reader can jump straight to chunk `[2]` to confirm the recommendation rather than taking the model's word for it.\n\n---\n\n# Exercise 2: Extract and Validate Citations\n\nGenerating citations is only useful if you can check them. After the model answers with bracketed chunk numbers, parse those citations out and verify each one points at a real chunk. This is the programmatic grounding check that catches a model citing a chunk that does not exist.\n\n### Requirements\n\n- Build numbered context and ask the model to answer with inline `[n]` citations.\n- Parse the cited numbers from the answer with a regex.\n- Print the answer, the set of cited chunk numbers, and whether every citation is a valid chunk index.\n\n\n**main.py**\n\n```python\nimport re\nfrom openai import OpenAI\n\nclient = OpenAI() # Uses OPENAI_API_KEY from environment\n\nchunks = [\n \"The Python GIL allows only one thread to execute Python bytecode at a time.\",\n \"For CPU-bound work, the multiprocessing module sidesteps the GIL with separate processes.\",\n \"For IO-bound work, threads still help because the GIL is released during blocking IO.\",\n]\n\nquestion = \"How should I speed up CPU-bound Python work?\"\n\ncontext = \"\\n\".join(f\"[{i + 1}] {chunk}\" for i, chunk in enumerate(chunks))\n\n# TODO: ask the model to answer using only the chunks, citing chunk numbers like [2]\n# TODO: parse the [n] citations from the answer with a regex\n# TODO: print the answer, the cited chunk numbers, and whether every citation is a valid chunk index\n```\n\n\n### Expected Output\n\nThe answer carries citations, and the validation confirms each cited number refers to a real chunk.\n\n\n```plaintext\nUse the multiprocessing module, which sidesteps the GIL with separate processes [2], because the GIL lets only one thread run Python bytecode at a time [1].\nCited chunks: [1, 2]\nAll citations valid: True\n```\n\n\n### Solution\n\n\n**main.py**\n\n```python\nimport re\nfrom openai import OpenAI\n\nclient = OpenAI() # Uses OPENAI_API_KEY from environment\n\nchunks = [\n \"The Python GIL allows only one thread to execute Python bytecode at a time.\",\n \"For CPU-bound work, the multiprocessing module sidesteps the GIL with separate processes.\",\n \"For IO-bound work, threads still help because the GIL is released during blocking IO.\",\n]\n\nquestion = \"How should I speed up CPU-bound Python work?\"\n\ncontext = \"\\n\".join(f\"[{i + 1}] {chunk}\" for i, chunk in enumerate(chunks))\n\nresponse = client.chat.completions.create(\n model=\"gpt-5.4-mini\",\n messages=[{\"role\": \"system\", \"content\": \"Answer using only the numbered chunks. Cite the chunk number in brackets after each claim.\"},\n {\"role\": \"user\", \"content\": f\"Context:\\n{context}\\n\\nQuestion: {question}\"}],\n)\nanswer = response.choices[0].message.content\n\ncited = sorted({int(n) for n in re.findall(r\"\\[(\\d+)\\]\", answer)})\nvalid = all(1 <= n <= len(chunks) for n in cited)\n\nprint(answer)\nprint(\"Cited chunks:\", cited)\nprint(\"All citations valid:\", valid)\n```\n\n\nThe regex `\\[(\\d+)\\]` pulls every bracketed number out of the answer, and checking each against the valid range turns citations from decoration into something you can audit automatically. A citation pointing at a nonexistent chunk is a clear hallucination signal worth flagging or rejecting. This parse-and-verify step is what lets you trust citations at scale instead of spot-checking them by hand.","pageType":"ai-engineering"}

Get Premium