An idempotence optimising framework for saving enterprise resources. This repo hosts the code for I-check with memory included: hosts a modular FastAPI backend that supports multiple intelligent agents with memory optimisation for cost and token savings. It uses document fingerprinting, hashing, and structured matching to avoid redundant LLM invocations.

# Step 1: Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Step 2: Install dependencies
pip install -r requirements.txt


# Step 3: Before running the server, create a `config.ini` file in the project root directory with the following structure:

[API_KEYS]
OPENAI_API_KEY = your_openai_api_key_here

[AWS]
aws_region = your_aws_region
aws_access_key_id = your_aws_access_key_id
aws_secret_access_key = your_aws_secret_access_key

[DB_DETAILS]
uri = your_mongodb_uri
db = your_database_name

# Step 3: Run the FastAPI server
python main.py

• FAQ Generator
• Summariser
• K_Search (Knowledge Search)
• Doc-Extractor

To ensure memory-based optimisation works as intended:

• Before calling any agent, call get_memory with required parameters. This returns whether a full, partial, or no match is found in memory.
• After getting a response from the agent, call save_memory to store the result in memory for future reuse.

Refer to this document for agent-wise details with examples: Implementation Doc

Each agent uses hierarchical memory checks based on the document hashes and specific agent parameters.

• Full Match: All focus areas covered with the required number of questions
• Partial Match: Fewer questions found for one or more focus areas
• No Match: No relevant questions found

• Full Match: Summary exists with exact focus areas and compression ratio
• Partial Match: Same focus areas but with a higher compression ratio; can be reused
• No Match: No relevant summary found

• Full Match: Exact query text and sources found
• Partial Match: Semantically similar queries found and reused
• No Match: No similar or exact queries found

• Full Match: All requested entity types found
• Partial Match: Some entity types found
• No Match: No matching entity types found

Each agent response includes an optional optimisation key if memory reuse was applied. This helps track cost and token savings.

"optimisation": {
    "original_tokens": 45539,
    "saved_tokens": 45539,
    "percent_saving_tokens": 100.0,
    "original_cost": 0.2163,
    "saved_cost": 0.2163,
    "percent_saving_cost": 100.0
}

Impact: The full agent response was reused from memory. No new tokens or cost were incurred.

"optimisation": {
    "original_tokens": 45539,
    "saved_tokens": 15179.67,
    "percent_saving_tokens": 33.33,
    "original_cost": 0.2163,
    "saved_cost": 0.0721,
    "percent_saving_cost": 33.34
}

Impact: Only a portion of the task was computed again; memory reuse covered the rest, resulting in significant savings.