text-summarizer-gi 0.2.0

Context-aware, LLM-powered rolling conversation summarizer

================================================================================

                      CONVERSATION SUMMARIZER

================================================================================



Progressive, context-aware conversation summarizer powered by Azure LLM.

Automatically compresses long conversations into structured summaries

while preserving user intent, key decisions, and action items.



================================================================================

                              QUICK START

================================================================================



1. INSTALL DEPENDENCIES

   ──────────────────────────────────────────────────────────────────────────

   pip install -r requirements.txt



   Or manually:

   pip install openai





2. SET UP AZURE CREDENTIALS

   ──────────────────────────────────────────────────────────────────────────

   The module needs three environment variables to call Azure OpenAI:



   AZURE_OPENAI_ENDPOINT       e.g. https://my-resource.openai.azure.com/

   AZURE_OPENAI_API_KEY        Your Azure API key

   AZURE_OPENAI_DEPLOYMENT     Your model deployment name



   Set them as follows:



   Windows (PowerShell):

      $env:AZURE_OPENAI_ENDPOINT="https://your-resource.openai.azure.com/"

      $env:AZURE_OPENAI_API_KEY="your-api-key-here"

      $env:AZURE_OPENAI_DEPLOYMENT="your-deployment-name"



   Windows (Command Prompt):

      set AZURE_OPENAI_ENDPOINT=https://your-resource.openai.azure.com/

      set AZURE_OPENAI_API_KEY=your-api-key-here

      set AZURE_OPENAI_DEPLOYMENT=your-deployment-name



   Mac/Linux:

      export AZURE_OPENAI_ENDPOINT="https://your-resource.openai.azure.com/"

      export AZURE_OPENAI_API_KEY="your-api-key-here"

      export AZURE_OPENAI_DEPLOYMENT="your-deployment-name"





3. RUN THE END-TO-END TEST

   ──────────────────────────────────────────────────────────────────────────

   This test reads sample_conversation.txt, calls your Azure LLM,

   and prints a formatted summary with sanity checks.



   From the project root:

      cd summarizer_package

      python test_summarizer.py



   Or from the workspace root (outer folder):

      python .\summarizer_package\test_summarizer.py



   Expected output:

      - Loads the sample conversation

      - Calls Azure LLM to summarize

      - Prints summary with overview, intent, decisions, and action items

      - Runs sanity checks (all should PASS)





4. RUN UNIT TESTS

   ──────────────────────────────────────────────────────────────────────────

   Unit tests use a mock LLM and do NOT require Azure credentials.



   From the project root:

      python -m pytest



   Or run test_conversation_summarizer.py directly:

      python test_conversation_summarizer.py



================================================================================

                           HOW TO USE IN CODE

================================================================================



from openai import AzureOpenAI

from conversation_summarizer import ConversationSummarizer, LLMClient

import os



# 1. Wire up your Azure client

azure_client = AzureOpenAI(

    azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],

    api_key=os.environ["AZURE_OPENAI_API_KEY"],

    api_version=os.environ.get("AZURE_OPENAI_API_VERSION", "2024-02-01"),

)



# 2. Create an LLMClient wrapper

llm_client = LLMClient(

    client=azure_client,

    model=os.environ["AZURE_OPENAI_DEPLOYMENT"]

)



# 3. Initialize the summarizer

summarizer = ConversationSummarizer(llm_client)



# 4. Progressive compression (auto-triggers when buffer hits threshold)

from conversation_summarizer import Message, Role

summarizer.add_message(Message(Role.USER, "Hello, I need help with X"))

summarizer.add_message(Message(Role.ASSISTANT, "Sure, tell me more..."))

# ... add more turns ...



# 5. Get context for the next LLM call

# (includes summary block + recent turns)

context = summarizer.get_context_messages()

context.append({"role": "user", "content": "What should we do next?"})

response = llm_client.complete(context)



# 6. Optionally summarize plain text directly

summary_text = summarizer.summarize_to_text(conversation_text)

print(summary_text)



================================================================================

                                 FILES

================================================================================



conversation_summarizer.py

      Core summarizer module. No Azure coupling—use any LLM backend

      by implementing the LLMClient interface.



test_summarizer.py

      End-to-end test: reads sample_conversation.txt, calls Azure LLM,

      prints summary and runs sanity checks.

      Requires: Azure credentials (see SETUP above).



test_conversation_summarizer.py

      Unit tests for ConversationSummarizer using a mock LLM.

      No Azure needed. Run with: pytest test_conversation_summarizer.py



sample_conversation.txt

      Sample conversation input. Edit with your own conversation

      using the format below.



requirements.txt

      Python package dependencies.



README.txt

      This file.



================================================================================

                          INPUT CONVERSATION FORMAT

================================================================================



Conversations should be plain text with a simple format:



   User: What do you need?

   Assistant: I need help building a leave management system.

   User: What are the requirements?

   Assistant: It should support multiple leave types, email notifications, ...



The summarizer treats "User:" and "Assistant:" as delimiters.

Edit sample_conversation.txt to test with your own sample.



================================================================================

                              CUSTOMIZATION

================================================================================



Auto-compress threshold:

   ConversationSummarizer(llm_client, auto_summarize_after=30)

   (compress after 30 messages instead of default)



Recent turns to keep in buffer:

   ConversationSummarizer(llm_client, keep_recent_turns=5)

   (retain 5 most recent messages after each compression)



Force compression:

   summary = summarizer.force_compress()

   (manually compress the entire buffer, regardless of size)



================================================================================