Skip to main content
In this tutorial, we’ll cover dynamic context injection—a technique that ensures your chatbot always provides accurate, up-to-date answers by pulling relevant information at runtime. You’ll start with a basic chat application using the OpenAI Chat Completion endpoint and evolve it into a robust system that sources context dynamically.

​
1. Basic Chat Completion (No External Context)

First, observe how GPT-3.5-turbo handles a query about the 95th Academy Awards (March 2023) without any added context. Because its training data cuts off in September 2021, it won’t know about later events. 
import os
import openai

openai.api_key = os.getenv("OPENAI_API_KEY")

response = openai.ChatCompletion.create(
    model="gpt-3.5-turbo",
    messages=[
        {
            "role": "system",
            "content": "You answer questions about the 95th Academy Awards held in March 2023. Answer only if you are sure."
        },
        {
            "role": "user",
            "content": "Which movie won the Best Picture award?"
        }
    ]
)

print(response.choices[0].message.content)
Expected response: 
I’m sorry, but I don’t have information on events after September 2021.
Without external context, the model defaults to stating its knowledge cutoff.

​
2. Injecting Static Context

To work around the cutoff, you can paste relevant excerpts from a trusted source directly into the prompt. For instance, from Good Morning America:
On Hollywood’s biggest night, Everything Everywhere All at Once reigned supreme, winning seven Oscars, including Best Picture.
Inject this snippet into your user message: 
import os
import openai

openai.api_key = os.getenv("OPENAI_API_KEY")

response = openai.ChatCompletion.create(
    model="gpt-3.5-turbo",
    messages=[
        {
            "role": "system",
            "content": "You answer questions about the 95th Academy Awards held in March 2023. Answer only if you are sure."
        },
        {
            "role": "user",
            "content": (
                "On Hollywood’s biggest night, \"Everything Everywhere All at Once\" reigned supreme, "
                "winning seven Oscars, including Best Picture. "
                "Which movie won the Best Picture award?"
            )
        }
    ]
)

print(response.choices[0].message.content)
Output: 
"Everything Everywhere All at Once" won the Best Picture award at the 95th Academy Awards.
Static context can quickly bloat your prompt and is tedious to maintain as information changes.

​
3. Why You Need Dynamic Context

Manual context updates are not scalable. Instead, build a pipeline that:
StepDescription
1. Store documentsSave articles, transcripts, or FAQs in a vector database
2. Embed user queriesGenerate embeddings for incoming questions
3. Retrieve relevant contextPerform a similarity search to fetch the most pertinent passages
4. Inject into the promptAppend retrieved snippets dynamically before calling the API
This dynamic context approach keeps your chatbot current without manual prompt edits.

​
4. Next Steps

In the following sections, we’ll demonstrate how to:
  • Generate embeddings for your documents
  • Set up and query a vector store (e.g., Pinecone, FAISS)
  • Compose the final prompt with retrieved context and call the Chat Completion endpoint
Stay tuned for the detailed implementation guide!

Watch Video

Practice Lab